DATA HUB

PROJECT #1

Sehat Watch · Portfolio Project

AI-Powered Pharmacovigilance Signal Intelligence

A small demonstration of how adverse-event report data can be turned into early safety signals and decisions.

Disclaimer. Demonstration project using synthetic data. Drug names are fictional. Nothing here is a clinical safety conclusion, medical advice or a statement about any real medicine.
224total reports
28report lines
26.3%serious
44.6%aged 65+

Chart: reported adverse events by seriousness

Dizziness
40
Nausea
37
Drowsiness
30
Hypoglycaemia
23
Gastric pain
22
GI bleeding
13
Dry cough
11
Headache
11
Hypotension
9
Diarrhoea
9
Confusion
8
Syncope
6
Rash
5
SeriousNon-serious

Reports by age group

18-44
40
45-64
84
65+
100

Reports by drug

Cardiovex
58
Glucoreg
63
Neurocalm
53
Painaid
50

3 key findings

  1. Most reported events: Dizziness (40), Nausea (37), Drowsiness (30) make up 47.8% of all 224 reports, and all are non-serious.
  2. Frequent is not the same as serious: 26.3% of reports are serious, and they come from only five event types, led by hypoglycaemia (23).
  3. Older patients carry the serious burden: patients aged 65+ account for 44.6% of reports but 93.2% of serious reports (55 of 59).

2 business implications

  1. Prioritise review by seriousness, not only volume. Dizziness leads on counts, but hypoglycaemia, GI bleeding, hypotension, confusion and syncope carry every serious report. Triage queues weighted by seriousness focus scarce safety-physician time where it matters.
  2. Segment signals by age group. Serious reports concentrate in patients aged 65+, so targeted monitoring, label review and risk-communication for older patients is a more efficient use of resources than a blanket approach.

Methodology

  1. Created 28 synthetic report lines for 4 fictional drugs (Cardiovex, Glucoreg, Neurocalm, Painaid) with Drug, Age Group, Adverse Event, Seriousness, Gender and Report Count.
  2. Cleaned and grouped events with consistent preferred-term-style names (MedDRA-style naming; no licensed dictionary used).
  3. Summed Report Count by event, seriousness, age group and drug; calculated each share of total reports.
  4. Ranked events by volume, then checked seriousness and age split to separate frequent from concerning.
  5. AI layer (proposed extension): NLP or an LLM could code free-text narratives to standard terms, deduplicate cases and pre-rank queues, with human pharmacovigilance review of every output. This demo does not run a model.

Limitations

  • Synthetic and small: no real-world reporting patterns, no denominators.
  • Spontaneous reports cannot show incidence or causality; underreporting and reporting bias are real.
  • No disproportionality statistics (PRR, ROR) and no case-level detail such as dose, duration or co-medication.
  • Counts are illustrative, so findings show the analytical approach only.

Project summary

AI-Powered Pharmacovigilance Signal Intelligence is a compact demonstration project from Sehat Watch showing how adverse-event reporting data can support drug-safety decisions. Using a synthetic dataset of 28 report lines across four fictional drugs, it identifies the most frequently reported events, separates frequent from serious events, and highlights how risk concentrates in older patients. It then outlines how AI-assisted coding and triage could speed up case processing while keeping qualified human review central. Built to showcase healthcare data analysis, pharmacovigilance understanding and practical business thinking, it is a learning demonstration and not a clinical safety assessment.

Synthetic dataset (28 rows)

DrugAge groupAdverse eventSeriousnessGenderReport count
Cardiovex65+DizzinessNon-seriousF14
Cardiovex65+HypotensionSeriousM9
Cardiovex45-64Dry coughNon-seriousF11
Cardiovex45-64DizzinessNon-seriousM8
Cardiovex65+SyncopeSeriousF6
Cardiovex18-44HeadacheNon-seriousF4
Cardiovex65+DizzinessNon-seriousM6
Glucoreg45-64NauseaNon-seriousF16
Glucoreg65+HypoglycaemiaSeriousM12
Glucoreg45-64DiarrhoeaNon-seriousM9
Glucoreg65+NauseaNon-seriousF10
Glucoreg18-44NauseaNon-seriousF5
Glucoreg65+HypoglycaemiaSeriousF7
Glucoreg45-64HypoglycaemiaSeriousF4
Neurocalm18-44DrowsinessNon-seriousF15
Neurocalm18-44DizzinessNon-seriousM7
Neurocalm45-64DrowsinessNon-seriousM9
Neurocalm65+ConfusionSeriousF8
Neurocalm65+DrowsinessNon-seriousF6
Neurocalm18-44HeadacheNon-seriousM3
Neurocalm45-64DizzinessNon-seriousF5
Painaid45-64Gastric painNon-seriousM13
Painaid65+GI bleedingSeriousM8
Painaid65+Gastric painNon-seriousF9
Painaid18-44NauseaNon-seriousF6
Painaid45-64RashNon-seriousF5
Painaid65+GI bleedingSeriousF5
Painaid45-64HeadacheNon-seriousF4

PROJECT #2

Sehat Watch · Portfolio Project

Clinical Data Quality & AI Analytics

Good AI starts with good data. A small demonstration of how to find data-quality problems in clinical records before they reach analytics or AI.

Disclaimer. Demonstration project using a fictional, deliberately imperfect dataset. No real patients. Nothing here is a clinical conclusion or medical advice.
40records checked
42%fully clean
97.1%field completeness
23issues found

Chart: records affected by issue type

Inconsistent category
9
Missing
7
Invalid range
5
Duplicate
2

Clean vs flagged records

Clean
17
Flagged
23

3 key findings

  1. Most records were not analysis-ready: only 17 of 40 records (42%) passed every check, even in a small and simple dataset.
  2. Inconsistent categories were the most common problem (9 records): variants such as "T2DM", "HTN", "Female" and "ACTIVE" would split groups and distort counts in any model or report.
  3. A few bad values change the story: the average glucose is 140 mg/dL on the raw data but 121 mg/dL after removing duplicates and out-of-range values, so decisions based on the raw average would be wrong.

Data-quality checks

  • Missing values: blank Age, Gender, Diagnosis, Glucose or Treatment (7 records).
  • Duplicates: repeated Patient IDs (2 records).
  • Invalid ranges: Age outside 0 to 110 or glucose outside 30 to 600 mg/dL (5 records, e.g. age 214, glucose 1050, glucose 0, and 5.6 which looks like a mmol/L value entered as mg/dL).
  • Inconsistent categories: values not matching the agreed list for Gender (F/M), Diagnosis and Treatment status (9 records).

2 business implications

  1. Make data quality a gate before analytics and AI. A simple quality scorecard checked before every model, report or trial analysis avoids wasted effort, misleading dashboards and unreliable AI outputs.
  2. Fix problems at the source. Entry-time validation, standard drop-down lists and shared coding rules reduce repeated clean-up costs and strengthen trust in hospital reporting, pharma real-world evidence and audit readiness.

Methodology

  1. Built a fictional 40-row clinic extract (age, gender, diagnosis, glucose lab result, treatment status) with deliberate errors.
  2. Defined four rule-based checks: missing values, duplicate IDs, invalid ranges and inconsistent categories.
  3. Flagged every record that failed at least one check and counted affected records per check.
  4. Compared a simple statistic (average glucose) before and after cleaning to show the impact on analysis.

AI / data-quality extension (proposed)

  • Category standardisation: fuzzy matching or an LLM maps variants ("T2DM", "Diabetes Type 2") to one standard term, with a human approving the mapping table.
  • Anomaly detection: a simple statistical or machine-learning model flags unusual values (like 1050 or a likely unit mix-up) for review.
  • Duplicate matching: similarity scoring finds near-duplicate patients that exact ID matching misses.
  • This demo uses plain rules only and does not run an AI model. Every AI suggestion would need human review.

Limitations

  • Fictional and small: real data has more complex errors and far more volume.
  • Range limits are simplified for demonstration and are not clinical reference ranges.
  • Rule checks find known problems only; they cannot confirm which value is correct.
  • No real patient data, no clinical validation, no regulatory or privacy review.

Project summary

Clinical Data Quality & AI Analytics is a compact Sehat Watch demonstration of why good AI starts with good data. Using a fictional, deliberately imperfect dataset of 40 patient records, it applies four data-quality checks (missing values, duplicates, invalid ranges and inconsistent categories), shows that only 42% of records were analysis-ready, and demonstrates how a few errors can distort a simple average. It closes with a practical AI-assisted extension for standardising categories and flagging anomalies, with human review at every step. It is a learning demonstration, not a clinical assessment.

Imperfect dataset (flagged rows highlighted)

IDAgeGenderDiagnosisGlucose (mg/dL)TreatmentIssue flagged
P00154FType 2 Diabetes142Active
P00261MHypertension98Active
P00347FemaleType 2 Diabetes156ActiveInconsistent category
P00439MDyslipidaemia92Completed
P00568FT2DM171ActiveInconsistent category
P00652MHypertension105activeInconsistent category
P00745FType 2 Diabetes138Discontinued
P008MHypertension101ActiveMissing
P00958FDyslipidaemiaActiveMissing
P01073maleHypertension110ActiveInconsistent category
P011214MType 2 Diabetes149ActiveInvalid range
P01244FHypertension5.6CompletedInvalid range
P01336FDyslipidaemia96Completed
P01463MDiabetes Type 2163ActiveInconsistent category
P01557FHypertension102ACTIVEInconsistent category
P01649MType 2 Diabetes0DiscontinuedInvalid range
P01766FDyslipidaemia99Active
P01841MHTN97CompletedInconsistent category
P01959FType 2 Diabetes152Active
P02072MHypertensionDiscontinuedMissing
P02150FType 2 Diabetes134ActvInconsistent category
P022-5MDyslipidaemia94ActiveInvalid range
P02364FHypertension108Completed
P02438MType 2 Diabetes1050ActiveInvalid range
P02555Hypertension103ActiveMissing
P02669FDyslipidaemia97Active
P02746MType 2 Diabetes145Completed
P02860F125ActiveMissing
P02953MHypertension100Discontinued
P03048FT2DM160ActiveInconsistent category
P03175MHypertension106Active
P03242FDyslipidaemia91Completed
P03367MType 2 Diabetes158Missing
P03456FHypertension104Active
P03562MDyslipidaemia95Discontinued
P03651FType 2 Diabetes141Active
P037FType 2 Diabetes147CompletedMissing
P03870MHypertension109Active
P00745FType 2 Diabetes138DiscontinuedDuplicate
P01959FType 2 Diabetes152ActiveDuplicate

PROJECT #3

Sehat Watch · Portfolio Project

Healthcare Analytics Dashboard

I built this hospital performance dashboard using a synthetic dataset modelled on a typical 300-bed multispecialty hospital in a North Indian city such as Lucknow, to show how basic KPIs support management decisions.

Disclaimer. Demonstration project using synthetic (simulated) data, modelled on typical ranges for North Indian private multispecialty hospitals: a monsoon fever and dengue surge, a winter respiratory peak and a 3 to 4-day average stay. It is not data from any real hospital and is not a clinical or performance assessment of any organisation.
80%avg bed occupancy
▲ 2.9 pts vs FY24-25
37 minavg waiting time
▼ 3 min vs FY24-25
3.9 daysavg length of stay
▼ 0.2 vs FY24-25
2,718ED visits / month
▲ 200 vs FY24-25
4.33 / 5patient satisfaction
▲ 0.01 vs FY24-25

Main chart: monthly bed occupancy (Apr 2023 to Mar 2026)

60%70%80%90%85% pressure lineApr 2023Apr 2024Apr 2025Mar 2026
Red dots mark the 3 months above 85% occupancy, when beds become hard to find for new admissions.

Comparison: avg waiting time (min)

FY 2023-24
42
FY 2024-25
40
FY 2025-26
37

Comparison: satisfaction (out of 5)

FY 2023-24
4.30
FY 2024-25
4.32
FY 2025-26
4.33

3 key findings

  1. Monsoon months are the pressure point: from July to October, ED visits run 12% higher and bed occupancy 8 points higher than other months. Occupancy peaked at 88.2% in Sep 2025, and 2 of the 3 months above 85% fall in the monsoon season.
  2. A process change cut waiting time even as volume grew: after a simulated triage redesign in Oct 2024, average waiting time fell from 43 to 36 minutes, while monthly ED visits rose from 2,352 to 2,661.
  3. Satisfaction follows crowding: months with occupancy above 80% average 4.26 satisfaction against 4.34 in other months, showing that patient experience dips when the hospital is stretched.

2 hospital management implications

  1. Plan for the monsoon surge in advance. Pre-schedule extra nursing and ED staff, hold back elective admissions and arrange surge beds from June, instead of reacting once occupancy passes 85%.
  2. Review waiting time, occupancy and satisfaction together each month. Seeing them side by side shows whether process changes such as triage redesign work and flags when capacity, not staff behaviour, is driving poor patient experience.

Methodology

  1. I built a synthetic monthly dataset of 36 observations (April 2023 to March 2026, Indian financial years) for a 300-bed hospital.
  2. I simulated seasonality (monsoon and winter surges), steady growth, and small random variation; occupancy is calculated from admissions, length of stay, days in month and bed count.
  3. I selected five KPIs and compared the latest financial year with the previous one.
  4. I used one trend chart and simple averages for comparisons. No advanced statistics are used.

Applying this to real hospital data

  • Obtain written permission from hospital management (and the ethics committee where required) before using any data.
  • Use only de-identified, monthly aggregate figures. No patient names, IDs or case-level records.
  • Follow India's data-protection requirements (Digital Personal Data Protection Act, 2023) and the hospital's own policies.
  • Agree KPI definitions with the hospital (for example how occupancy and waiting time are measured) and run data-quality checks first.
  • Document the source, period and approvals on the page.

Project summary

Healthcare Analytics Dashboard is a compact Sehat Watch project that I designed and built to demonstrate hospital performance analytics. Using three years of synthetic monthly data modelled on a 300-bed North Indian multispecialty hospital, it tracks bed occupancy, waiting time, length of stay, emergency visits and patient satisfaction, and turns them into clear findings: monsoon-season capacity pressure, the effect of a triage redesign, and the link between crowding and patient experience. It shows healthcare data visualisation, KPI thinking and decision support in a form managers can act on. It is a learning project using synthetic data, not an assessment of any real hospital.

Clean dataset (36 months; months above 85% occupancy highlighted)

MonthAdmissionsED visitsAvg wait (min)ALOS (days)Bed occupancy (%)Satisfaction (/5)
Apr 20231,4402,063364.267.24.38
May 20231,4101,990344.263.74.4
Jun 20231,4192,036384.266.24.32
Jul 20231,5862,392434.373.34.29
Aug 20231,7082,454474.379.04.22
Sep 20231,6932,565484.380.94.23
Oct 20231,6812,475434.275.94.26
Nov 20231,4832,180384.167.64.36
Dec 20231,6172,357434.273.04.27
Jan 20241,6682,461454.275.34.25
Feb 20241,5802,196444.174.54.25
Mar 20241,5772,232404.169.54.34
Apr 20241,6132,280434.173.54.3
May 20241,5502,266394.168.34.36
Jun 20241,5522,298424.170.74.33
Jul 20241,7392,585454.278.54.27
Aug 20241,8082,692494.179.74.19
Sep 20241,8592,813524.286.84.11
Oct 20241,8282,771404.282.64.29
Nov 20241,6312,340344.174.34.37
Dec 20241,7432,561313.973.14.45
Jan 20251,7952,612344.077.24.41
Feb 20251,7192,461394.081.94.34
Mar 20251,6862,536324.072.54.45
Apr 20251,7212,491333.974.64.38
May 20251,6592,387283.969.64.44
Jun 20251,6822,428313.972.94.4
Jul 20251,9332,838373.981.14.34
Aug 20251,9772,922424.085.04.25
Sep 20252,0353,004433.988.24.19
Oct 20252,0042,898423.984.04.27
Nov 20251,7482,558323.873.84.46
Dec 20251,9402,838363.879.34.35
Jan 20261,9822,903393.983.14.29
Feb 20261,8492,700433.985.84.26
Mar 20261,8742,651343.876.64.36

PROJECT #4

Sehat Watch · Portfolio Project

MedTech Market Intelligence

I built this competitive analysis of the remote patient monitoring (RPM) market to show how data can support healthcare business strategy. It compares 9 illustrative companies on customer, technology, pricing and positioning.

Disclaimer. Demonstration project. All companies are fictional and any resemblance to real companies is coincidental. Prices and the clinical depth scores are my own illustrative assumptions modelled on the general shape of the RPM market. This is not real market data, investment advice or a claim about any real company.

Competitive-positioning chart: clinical depth vs price

₹0₹1000₹2000₹30000246810Affordable clinical-grade zoneCardioLink HealthGlucoSense CloudVitalNestHomeWard AICareLoopPulseBridgeMediTrack RuralWellBandClinicaHubClinical depth score (1 = wellness, 10 = clinical-grade)
Price = illustrative ₹ per patient per month. Colour = main target customer. HospitalsClinics & payersPublic healthConsumers

3 market insights

  1. Price follows clinical depth: the 3 most clinical-grade players (depth 7 or more) average about ₹1,833 per patient per month, against ₹425 for the other 6.
  2. Affordable clinical-grade is thinly served: only 2 of 9 companies (MediTrack Rural, ClinicaHub) offer mid-to-high clinical depth at or below ₹500, and both focus on narrow segments.
  3. Recurring revenue is the norm: 7 of 9 companies use subscriptions or licences, and only one relies mainly on selling devices. Competitive advantage comes from workflow, integration and evidence rather than hardware.

2 strategic implications

  1. Target the white space. A newcomer or partner could offer affordable, clinically credible RPM for tier-2 and tier-3 city hospitals, where premium platforms are too costly and consumer wearables are not credible enough.
  2. Compete on integration and outcomes. Hospital buyers reward EHR/HMIS integration and proof of results such as fewer readmissions, so a defensible strategy builds those proof points instead of competing on device price.

Competitor matrix

CompanyAI alertsEHR linkHome useLow-bandwidthRecurring fee
CardioLink Health●●●○●
GlucoSense Cloud●●●○●
VitalNest○●●○●
HomeWard AI●●●○●
CareLoop●○●○●
PulseBridge○○●○●
MediTrack Rural○○●●○
WellBand○○●○○
ClinicaHub○●○●●
● offered, ○ not offered (illustrative assessment)

Methodology

  1. I chose one focused segment, remote patient monitoring, and created 9 fictional companies that reflect typical market archetypes (cardiac, diabetes, hospital-at-home, consumer, rural, device-agnostic).
  2. I defined comparison variables: category, target customer, key technology, positioning, pricing model and competitive advantage.
  3. I assigned each company a clinical depth score and an illustrative price per patient per month, then plotted them to compare positioning.
  4. I built a capability matrix and counted patterns to draw 3 insights and 2 strategic implications.
  5. To turn this into a real analysis, I would replace the fictional entries with companies sourced from public information (websites, filings, regulatory registers) and cite each fact.

Limitations

  • Fictional companies and illustrative numbers, so no real market sizes, shares or prices.
  • Clinical depth scores are subjective analyst judgements, not measured values.
  • A small sample of 9 cannot represent the whole market; regulation, reimbursement and data privacy are not analysed.
  • Insights show the analytical approach and are not investment or business advice.

Project summary

MedTech Market Intelligence is a compact Sehat Watch project that I designed to demonstrate data-driven healthcare business strategy. It compares 9 illustrative remote patient monitoring companies on target customer, technology, pricing and positioning, maps them on a clinical-depth versus price chart, and identifies a thinly served space for affordable, clinical-grade monitoring. It closes with two strategic implications and a plan for replacing the fictional data with publicly sourced, cited information. It is a learning project using fictional companies, not a real market assessment.

Company comparison dataset

CompanyProduct categoryTarget customerKey technologyPositioningPricing / business modelCompetitive advantage
CardioLink HealthCardiac monitoringHospitals and cardiology unitsECG patch + cloud arrhythmia detectionPremium, clinical-gradePer-patient monthly fee + device leaseCardiologist-reviewed signals
GlucoSense CloudDiabetes monitoringDiabetes clinics and insurersCGM data integration + AI trend alertsUpper-mid, specialistSaaS fee per clinic + per patientDiabetes coaching workflow
VitalNestHome vitals kitHospitals (post-discharge care)Bluetooth BP/SpO2/temperature devices + appMid-marketKit rental + monthly feePost-discharge readmission programmes
HomeWard AIHospital-at-home platformLarge hospital chainsAI risk scoring + deep EHR integrationPremium enterpriseAnnual enterprise licenceDeep EHR/HMIS integration
CareLoopChronic-care coachingEmployers and insurersMobile app + coaches + simple devicesWellness-clinical hybridPer member per monthEngagement and behaviour change
PulseBridgeElder-care monitoringFamilies (consumers)Smartwatch + fall detectionConsumer, simpleDirect-to-consumer subscriptionFamily alerts and simple UX
MediTrack RuralPrimary-care RPMGovernment programmes and NGOsOffline-first app + low-cost devicesAffordable, volumePer-project contractWorks in low connectivity
WellBandConsumer wearableFitness and wellness consumersOptical sensors + wellness insightsLow-cost consumerDevice sales + premium appBrand and price
ClinicaHubDevice-agnostic hubSmall clinics and GPsMulti-device data aggregation dashboardValue, vendor-neutralPer-clinic subscriptionWorks with any device

PROJECT#5

Sehat Watch · Portfolio Project

AI in Healthcare: Use-Case & ROI Analysis

I built this simple framework to compare 9 healthcare AI use cases and decide where an organisation should start. It scores each on six criteria, plots them on a 2×2 matrix and works through one transparent ROI example.

Disclaimer. Illustrative demonstration. The scores are my own judgement (1 to 5), not measured evidence, and the ROI calculation is a worked example based on clearly labelled assumptions. It does not report actual ROI achieved by any organisation or product and is not a guarantee of results.

2×2 prioritisation matrix: value vs feasibility

QUICK WINSSTRATEGIC BETSFILL-INSMONITORFeasibility (1 = hard, 5 = easy) →Value (1 = low, 5 = high) →123456789
1 Clinical documentation (ambient AI scribe)
2 Administrative automation (billing and claims coding)
3 Pharmacovigilance case intake and signal detection
4 Medical imaging triage (radiology)
5 Patient triage (symptom and ED triage)
6 Drug discovery (molecule screening)
7 Appointment no-show prediction
8 Readmission risk prediction
9 Clinical trial patient matching

Scoring table (1 to 5)

#Use caseImpactComplexityData readyAI maturityROI potentialRiskValueFeasibilityQuadrant
1Clinical documentation (ambient AI scribe)5234434.53.50Quick win
2Administrative automation (billing and claims coding)4244524.54.00Quick win
3Pharmacovigilance case intake and signal detection4344434.03.50Quick win
4Medical imaging triage (radiology)5434444.52.75Strategic bet
5Patient triage (symptom and ED triage)4423353.52.00Strategic bet
6Drug discovery (molecule screening)5523233.52.25Strategic bet
7Appointment no-show prediction3144313.04.50Fill-in
8Readmission risk prediction3334333.03.25Monitor
9Clinical trial patient matching3333333.03.00Monitor
Value = average of Impact and ROI potential. Feasibility = average of Data readiness, AI maturity, (6 − Complexity) and (6 − Risk). High value and feasibility are 3.5 or more.

3 key findings

  1. The best starting points are workflow-heavy and data-ready: the 3 quick wins (clinical documentation, administrative automation, pharmacovigilance case intake and signal detection) combine strong value with ready data and moderate risk.
  2. Clinical-facing AI carries the most risk: clinical-care use cases average a risk score of 4.0, against 2.0 for operations, and also score lower on data readiness (2.7 vs 3.7).
  3. High impact does not mean easy: 3 high-value use cases (medical imaging triage, patient triage, drug discovery) are strategic bets because feasibility is low, so they need stronger data, validation and governance first.

2 strategic recommendations

  1. Start with 2 or 3 quick wins as measured pilots. Set a baseline, one success metric and a fixed pilot period for each, so the organisation gains evidence of benefit before scaling.
  2. Build the foundation for the strategic bets. Invest in data quality, clinical validation, human oversight and governance now, so higher-risk use cases like imaging and triage are ready when pilots succeed.

Worked ROI example: ambient clinical documentation

Assumptions (all illustrative, to be replaced with the organisation's own figures): 100 doctors, 250 working days a year, value of clinician time ₹1,500 per hour, licence ₹8,000 per doctor per month, one-time implementation ₹15 lakh and first-year training and admin ₹5 lakh. Year-1 cost: ₹1.16 crore.

Scenario (assumed time saved)Value of timeYear-1 costNetROI
20 min / doctor / day₹1.25 crore₹1.16 crore₹0.09 crore8%
40 min / doctor / day₹2.50 crore₹1.16 crore₹1.34 crore116%
60 min / doctor / day₹3.75 crore₹1.16 crore₹2.59 crore223%

Break-even: under these assumptions the tool pays back at about 19 minutes saved per doctor per day. Time saved is not cash saved unless it is turned into extra patient capacity or reduced overtime, so the real benefit must be measured in a pilot. ROI = (value − cost) ÷ cost.

Methodology

  1. I selected 9 common healthcare AI use cases across care delivery, operations and life sciences.
  2. I scored each from 1 to 5 on business impact, implementation complexity, data readiness, AI maturity, potential ROI and risk.
  3. I combined the scores into a value score and a feasibility score, then plotted them on a 2×2 matrix with a 3.5 cut-off.
  4. I built one worked ROI example with explicit assumptions and three scenarios.
  5. With a real client, I would score together with stakeholders, use their data for the assumptions and test the ranking with a pilot.

Risks and limitations

  • Scores are subjective and equally weighted, and a different weighting could change the ranking.
  • The ROI example uses assumed figures, not observed results, and ignores ongoing costs such as monitoring and change management.
  • Use cases are generic. Real choices depend on the organisation, local data, regulation and clinical safety review.
  • AI risks such as bias, privacy, errors and over-reliance need governance and human oversight in practice.

Project summary

AI in Healthcare: Use-Case & ROI Analysis is a compact Sehat Watch project that I designed to show how healthcare organisations can prioritise AI investments. It scores 9 use cases on impact, complexity, data readiness, AI maturity, ROI potential and risk, plots them on a value-versus-feasibility matrix, and works through a transparent ROI example with labelled assumptions and a break-even point. The results point to starting with workflow-focused quick wins while building the data and governance foundations for higher-risk clinical applications. It is an illustrative framework, not evidence of real-world ROI.

PROJECT#6

Sehat Watch · Portfolio Project

Digital Health Business Case Lab

I built this simple business case for a fictional remote patient monitoring (RPM) platform for chronic disease care, with a financial model small enough to explain line by line.

Disclaimer. Illustrative demonstration. The company, customers and every number are fictional assumptions that I chose for learning. They are not market data, forecasts or financial advice, and nothing here has been validated with real hospitals.

Financial chart: revenue vs total cost (₹ lakh)

01002003003674Year 1Net −₹38.4 lakh120133Year 2Net −₹13.0 lakh270218Year 3Net ₹52.0 lakh
RevenueTotal cost (variable + fixed). 1 crore = 100 lakh.

Problem

People with type 2 diabetes and hypertension often see a doctor only every few months. Between visits, hospitals have little visibility into blood pressure and glucose, so problems are caught late and follow-up depends on the patient remembering to return.

Target users and proposed solution

Users: adult patients with diabetes or hypertension (patients), doctors and nurse care coordinators (clinical users), and hospital administrators (buyers) at private hospitals in tier-2 Indian cities.

Solution (fictional "ChronicCare Connect"): a home kit with a blood pressure monitor and glucometer that sends readings to a simple phone app, with nurse-led monitoring, alerts to doctors for out-of-range readings and a monthly progress report for the hospital.

Assumptions and revenue model

  • Customers: hospitals sign up gradually: 4 in Year 1, 10 in Year 2 and 18 in Year 3.
  • Patients: average enrolled patients per hospital of 150, 200 and 250 in Years 1 to 3.
  • Revenue model: the hospital pays ₹500 per enrolled patient per month (a subscription), and may include it in a chronic-care package.
  • Variable cost: ₹200 per patient per month: device kit cost spread over its life (₹80), cloud and messaging (₹20) and nurse monitoring and support (₹100).
  • Fixed costs: team, compliance and legal, sales and tools: ₹60.0 lakh, ₹85.0 lakh and ₹110.0 lakh in Years 1 to 3.
  • Formula: revenue = patients × ₹500 × 12. Contribution = ₹300 per patient per month (₹3,600 a year).

3-year projection (₹ lakh)

Year 1Year 2Year 3
Hospital partners41018
Avg enrolled patients6002,0004,500
Revenue₹36.0 lakh₹120.0 lakh₹270.0 lakh
Variable costs₹14.4 lakh₹48.0 lakh₹108.0 lakh
Gross profit₹21.6 lakh₹72.0 lakh₹162.0 lakh
Fixed costs₹60.0 lakh₹85.0 lakh₹110.0 lakh
Net profit−₹38.4 lakh−₹13.0 lakh₹52.0 lakh
Cumulative profit−₹38.4 lakh−₹51.4 lakh₹0.6 lakh

Break-even estimate

At the Year 3 fixed-cost level of ₹110.0 lakh, the business breaks even when it has about 3,056 average enrolled patients (fixed costs ÷ yearly contribution per patient = 11,000,000 ÷ 3,600). The plan reaches 4,500 in Year 3, so annual profit turns positive in Year 3. Cumulative profit is roughly zero at the end of Year 3, with a lowest point of −₹51.4 lakh.

ScenarioYear 3 profitCumulative, Year 3
Base case₹52.0 lakh₹0.6 lakh
Enrolment 30% lower₹3.4 lakh−₹76.1 lakh
Fee ₹100 lower (₹400)−₹2.0 lakh−₹84.6 lakh

3 business insights

  1. Unit economics are healthy, but only at scale: each patient contributes ₹300 a month (60% of the fee), yet fixed costs of ₹60.0 lakh in Year 1 mean losses until enrolment grows.
  2. Funding need peaks at about ₹51.4 lakh: the cumulative loss is lowest at the end of Year 2, so with a 20% buffer the plan needs roughly ₹61.7 lakh of funding.
  3. The case is sensitive: with 30% fewer enrolled patients or a ₹100 lower fee, Year 3 profit falls from ₹52.0 lakh to ₹3.4 lakh and −₹2.0 lakh, so enrolment and pricing are the numbers to test first.

2 strategic recommendations

  1. Validate with a small pilot first. Run 2 to 3 hospitals before building more, to test real enrolment per hospital, the willingness to pay ₹500 and patient drop-out.
  2. Plan funding for the dip and track two lead metrics. Raise about ₹61.7 lakh, and track enrolled patients per hospital and 6-month retention, since they drive every line of the model.

Key risks

  • Adoption: hospitals may be slow to sign or to enrol patients.
  • Engagement: patients may stop taking readings, reducing value.
  • Regulation and privacy: health data (India's Digital Personal Data Protection Act, 2023), device rules and clinical oversight must be handled properly.
  • Costs and pricing: device costs, nurse time and price pressure could narrow margins.

Methodology

  1. I chose a focused problem and defined users, solution and revenue model.
  2. I set explicit assumptions for hospitals, enrolled patients, price and costs.
  3. I calculated revenue, variable costs, fixed costs and profit for three years, and a break-even point from the contribution per patient.
  4. I tested two simple scenarios (lower enrolment, lower price) and drew insights and recommendations.
  5. With a real venture I would replace each assumption with pilot or market evidence.

Assumptions and limitations

  • All figures are assumptions, not market research, and no real customers or prices were tested.
  • The model is simplified: no tax, inflation, discounting, working capital, or staged hiring detail.
  • Average enrolled patients per year hides month-by-month ramp-up and drop-out.
  • It is a financial planning demonstration, not a clinical, regulatory or investment assessment.

Project summary

Digital Health Business Case Lab is a compact Sehat Watch project that I designed to practise healthcare business planning. It builds a simple three-year business case for a fictional remote patient monitoring platform for chronic disease care, with transparent assumptions, a revenue and cost projection, a break-even estimate and scenario tests. It shows that the model works only at sufficient scale, estimates the funding needed, and recommends a small pilot to validate enrolment and pricing. It is an illustrative learning project using fictional assumptions, not a forecast.

PROJECT #7

Sehat Watch · Portfolio Project

Healthcare Patient Journey Analytics

I built this analysis of 40 fictional patient journeys to show where patients wait, where they drop off and what that means for hospital operations.

Disclaimer. Demonstration project using fictional, synthetic data. The patients, times and scores were simulated by me for learning. They are not from any real hospital and are not a clinical or performance assessment.

Patient journey visualization

Appointment40Stage 1100% remain›2 leftRegistration38Stage 295% remain›1 leftWaiting37Stage 392% remain›4 leftConsultation33Stage 482% remain›4 leftDiagnostictest29Stage 572% remain›6 leftTreatment23Stage 657% remain›5 leftFollowup18Stage 745% remainNumbers show patients reaching each stage. "left" shows patients who did not continue to the next stage.

Biggest bottleneck: average minutes in each stage

Registration
12
Waiting
57
Consultation
13
Diagnostic test
70
Treatment
42
Follow up
11
Red marks the longest stage. The appointment stage is excluded because it is measured in days, not minutes (average booking lag about 3 days).

3 key findings

  1. Fewer than half completed the journey: only 18 of 40 patients (45%) reached follow up. They averaged 2.8 visits and 23 days from first appointment to follow up.
  2. Diagnostic testing is the biggest bottleneck: it is the longest stage at 70 minutes on average and the point where the most patients are lost (6 of 40, or 15%).
  3. Long waits and drop off go with lower satisfaction: patients waiting over 60 minutes before consultation rated their experience 3.5 out of 5, against 3.7 for shorter waits, and patients who dropped off averaged 3.5 against 3.8 for those who completed.

2 operational and business implications

  1. Diagnostics capacity shapes the whole journey. Queues and slow reporting delay treatment and cause repeat visits, so fixing the test stage helps patient flow across the hospital.
  2. Every lost patient is a continuity and revenue loss. Patients who leave before treatment or follow up reduce hospital income and weaken care outcomes, so retention across stages is a core business metric.

2 recommendations to improve the patient journey

  1. Book diagnostic tests before the patient leaves consultation, offer fixed time slots and send results digitally, to cut queue time and extra visits.
  2. Use time slots and SMS or WhatsApp updates to reduce waiting and send reminders for treatment and follow up, so fewer patients drop off between stages.

Methodology

  1. I defined seven stages: appointment, registration, waiting, consultation, diagnostic test, treatment and follow up.
  2. I simulated 40 patient journeys with the last stage each patient reached, minutes spent in each stage, number of visits, journey length in days and a satisfaction score from 1 to 5.
  3. I counted how many patients reached each stage to find drop off points, and averaged minutes per stage to find the bottleneck.
  4. I compared satisfaction across waiting time groups and between completed and dropped journeys.
  5. With real data, I would use de identified hospital system timestamps instead of simulated values.

Limitations

  • The data is fictional and small, so findings show the method and are not real evidence.
  • Satisfaction depends on the rules I used to simulate it, so the link to waiting time is built in by design.
  • Real patients may skip stages, return later or move between hospitals, which this simple flow does not capture.
  • Reasons for dropping off are not recorded and would need patient feedback or follow up calls.

Project summary

Healthcare Patient Journey Analytics is a compact Sehat Watch project that I designed to show how patient flow data can guide hospital improvement. Using 40 fictional patient journeys across seven stages, it maps where patients wait and drop off, identifies diagnostic testing as the biggest bottleneck, and links long waits and drop off to lower satisfaction. It ends with two operational implications and two practical recommendations. It is an illustrative learning project using synthetic data, not a real hospital assessment.

Patient journey dataset (40 journeys, incomplete journeys highlighted)

IDDepartmentLast stage reachedVisitsWaiting (min)Diagnostic (min)Journey (days)Satisfaction (1 to 5)
J01OrthopaedicsTreatment35270103.3
J02CardiologyConsultation15143.8
J03OrthopaedicsFollow up46372233.7
J04MedicineDiagnostic test2437953.3
J05CardiologyFollow up45673243.7
J06MedicineTreatment3552963.6
J07GynaecologyWaiting16333.3
J08OrthopaedicsConsultation16223.3
J09OrthopaedicsFollow up45663203.9
J10MedicineFollow up445100223.8
J11MedicineFollow up466103263.2
J12OrthopaedicsDiagnostic test2674753.2
J13CardiologyFollow up44978223.9
J14MedicineConsultation15223.7
J15MedicineDiagnostic test2416883.4
J16OrthopaedicsAppointment144.0
J17CardiologyFollow up46139233.9
J18MedicineFollow up48039263.7
J19MedicineFollow up474111273.1
J20MedicineAppointment144.2
J21OrthopaedicsFollow up453120223.4
J22GynaecologyTreatment35191103.3
J23OrthopaedicsRegistration144.1
J24OrthopaedicsWaiting16643.5
J25GynaecologyConsultation16723.4
J26CardiologyDiagnostic test2487863.4
J27MedicineWaiting13444.0
J28GynaecologyWaiting14443.6
J29MedicineFollow up44966204.2
J30MedicineTreatment3624083.3
J31OrthopaedicsTreatment35684113.3
J32MedicineFollow up47165213.7
J33CardiologyFollow up46945203.8
J34MedicineDiagnostic test2506263.6
J35OrthopaedicsDiagnostic test2607963.3
J36CardiologyFollow up45058234.3
J37GynaecologyFollow up44967244.1
J38CardiologyFollow up46935273.9
J39OrthopaedicsFollow up45167234.0
J40CardiologyFollow up46396223.5

PROJECT #8

Remote Patient Monitoring Report
Sehat Watch · Healthcare Industry Intelligence Report

Remote Patient Monitoring: India's rails are ready, the services are not

A compact research brief on how connected monitoring could close India's chronic care gap, with a focus on Lucknow and Uttar Pradesh.

By Yashi Srivastava Lucknow, Uttar Pradesh July 2026 Trend: Remote Patient Monitoring (RPM)
Summary

I looked at how remote patient monitoring could close India's chronic care gap, where 101 million people live with diabetes and fewer than 7% reach full disease control. The report tracks a global market expected to roughly double to about USD 57 billion by 2030, India's 96 crore ABHA health IDs, and local anchors like SGPGI's tele-ICU network. Key finding: India's digital health rails are ready, but paid home-monitoring services are not, and Uttar Pradesh is where that gap matters most.

Dashboard snapshot

ABHA health IDs grew to 96 crore in under five years

Cumulative ABHA IDs created, in crore. About 6.5x since 2021.

2021: 14.7 crore 2022: 30.4 crore 2023: 50.6 crore 2024: 72.2 crore 2025: 84.5 crore July 2026: 95.99 crore 14.7 30.4 50.6 72.2 84.5 96.0 2021 2022 2023 2024 2025 Jul 2026

Source: NHA / MoHFW. Calendar-year totals 2021 to 2025; July 2026 figure from Lok Sabha reply.

101M
Indians living with diabetes (ICMR-INDIAB)
<7%
of self-reported diabetics at full ABC control
15.3 cr
ABHA accounts in Uttar Pradesh, the highest of any state

Market and problem overview

Remote patient monitoring uses connected devices (BP cuffs, glucometers, ECG patches, pulse oximeters, under-mattress sensors) to send patient vitals to clinicians outside the ward or clinic. The global market is valued at roughly USD 26 to 28 billion today, and most analyst houses expect it to at least double by 2030.

The problem RPM targets is simple. India's chronic disease load is huge, specialists are concentrated in metros and medical colleges, and most deterioration in a diabetic or cardiac patient happens at home, between visits, where nobody is watching. RPM turns that gap into a data stream a care team can act on.

Why the trend matters

  • Chronic disease is a monitoring problem. Diabetes and hypertension need daily readings, but most patients see a doctor once every few months.
  • India now has the digital rails. ABHA health IDs, linked records and the eSanjeevani telemedicine network give RPM data somewhere to go.
  • Uttar Pradesh is a rising-wave state. ICMR-INDIAB recorded UP with the lowest diabetes prevalence in the country (4.8%), and lower-HDI states showed more prediabetes than diabetes. The wave is still building, so the window for preventive monitoring is open now.
  • Hospitals are short on nurses. Continuous ward monitoring lets one nurse station watch many beds and catch deterioration earlier.

Key data points

#Data pointSource
1Global RPM market: USD 27.72B (2024) to USD 56.94B (2030), 12.7% CAGRMarketsandMarkets
22025 sizing varies almost 3x between firms (USD 9.4B vs USD 26.0B) depending on how RPM is definedResearchAndMarkets
3India: 101M people with diabetes, 136M with prediabetes, 315M with hypertensionICMR-INDIAB
4Under 7% of self-reported diabetics reach combined ABC (HbA1c, BP, cholesterol) controlNMJI
595.99 crore ABHA IDs created nationally as of 31 July 2026Lok Sabha reply
6Uttar Pradesh leads all states with 15.3 crore+ ABHA accountsIndia Med Today
7US Medicare added CPT 99445 in 2026, paying for just 2 to 15 days of monitoring (previously a 16-day minimum)Physicians Practice

Key players

Global device majors

Philips, Medtronic, Masimo, Siemens Healthineers

Named among leading RPM players in major market research coverage.

Glucose monitoring

Abbott, Dexcom

Continuous glucose monitors are the most common home RPM device for diabetes.

Indian startup

Dozee, Bengaluru

Contactless under-mattress sensor; reports 300+ hospitals and 16,000+ monitored beds across 4 countries.

Public platform

eSanjeevani (MoHFW)

National telemedicine service; UP alone crossed 50 lakh consultations by early 2023.

Lucknow anchors

SGPGIMS and KGMU

SGPGI runs a tele-ICU hub linking district medical college ICUs; KGMU offers eSanjeevani OPD.

Technology developments

  • Contactless sensing. Ballistocardiography sheets read heart rate, breathing and movement from micro-vibrations under the mattress, with no wires on the patient.
  • AI early warning. Vendors now sell deterioration prediction, not just raw vitals. Dozee claims its system can flag deterioration up to 16 hours ahead (a company claim).
  • Interoperability. ABDM uses FHIR standards and consent-based record sharing; over 105 crore health records were linked by July 2026.
  • Short-episode reimbursement. The US 2-to-15-day code makes post-discharge and post-surgery monitoring billable, a model Indian insurers can study.

Opportunities and risks

Opportunities
  • Home BP and glucose programs for India's 136M prediabetic adults, before they convert
  • Hub-and-spoke RPM in UP: Lucknow tertiary centres watching patients in district hospitals and Ayushman Arogya Mandirs
  • Post-discharge monitoring to cut readmissions in private hospital chains
  • RPM readings plugged into ABHA-linked records for continuity of care
Risks
  • Data privacy: DPDP Rules notified Nov 2025, with full obligations and penalties from May 2027
  • No Indian reimbursement code for RPM yet, so most programs are out-of-pocket or hospital-funded
  • Device accuracy and alarm fatigue: too many false alerts and nurses stop responding
  • Digital divide: patchy connectivity and low device literacy among elderly rural patients

Future outlook

Over the next three to five years, RPM in India is likely to move from ICU and ward pilots toward home chronic care. Three things will decide the pace: a reimbursement pathway (PM-JAY or private insurers paying for monitoring days), DPDP-ready consent design, and integration with ABHA records so readings follow the patient. Globally, analyst forecasts range from 12.7% to 25% annual growth, with AI triage becoming the main differentiator rather than the hardware.

Insights and strategic implications

Three major insights
  1. The infrastructure is ahead of the use case. India has 96 crore health IDs but almost no paid home-monitoring programs.
  2. The control gap is the real market. With under 7% of diabetics at full ABC control, RPM's value is continuous follow-up between visits, not detection.
  3. Market size headlines are unreliable. 2025 estimates differ by nearly 3x because firms define RPM differently. Any pitch should state which definition it uses.
Two strategic implications

For hospitals in Lucknow

Start with post-discharge cardiac and diabetic patients from SGPGI, KGMU or large private chains, where readmission costs justify monitoring even without insurance cover.

For health-tech builders

Design for ABHA linking and DPDP consent from day one. Compliance before the May 2027 deadline is a sales advantage with hospitals and insurers.

Research methodology and limitations

Desk research carried out from Lucknow in July 2026. I shortlisted four trends (AI in healthcare, digital therapeutics, RPM, health data platforms) and picked RPM because it had the clearest link to India's chronic disease data and to public infrastructure in UP. Sources were limited to government releases and parliamentary replies, peer-reviewed studies, market research summaries and company disclosures. Each figure was cross-checked against a second source where possible, and the chart uses official ABDM figures only.

Limitations
  • Market sizes come from free press summaries, not full paid reports, and firms use different definitions.
  • Company figures (Dozee hospitals, beds, 16-hour prediction) are self-reported.
  • ABHA counts measure IDs created, not active use.
  • No primary interviews with Lucknow clinicians or patients were done for this version.
  • ICMR-INDIAB data was collected between 2008 and 2020, so current prevalence is likely higher.

Sources

  1. MarketsandMarkets, RPM Market Global Forecast to 2030 (press summary)
  2. ResearchAndMarkets, RPM Market Industry Report 2025 to 2035 (BusinessWire)
  3. ResearchAndMarkets, RPM System Market Size report
  4. ICMR-INDIAB findings, Lancet Diabetes and Endocrinology (National Herald)
  5. Lessons learnt from the ICMR-INDIAB study, National Medical Journal of India
  6. ABHA count, Lok Sabha reply of 31 July 2026 (Kashmir Life)
  7. ABDM crosses 90 crore ABHA accounts (India Med Today)
  8. ABDM Mission Steering Group, July 2026 (Indian Pharma Post)
  9. eSanjeevani in Uttar Pradesh, NHM UP presentation (WHO SEARO)
  10. SGPGIMS Lucknow tele-ICU (ANI)
  11. Dozee company profile
  12. DPDP Act and Rules implementation timeline
  13. 2026 Physician Fee Schedule final rule and RPM (Physicians Practice)
Sehat Watch · Healthcare Industry Intelligence Report · Yashi Srivastava · Lucknow · July 2026

PROJECT#9

Predictive Healthcare Analytics
Sehat Watch · Analytics Project

Predictive Healthcare Analytics: a readmission risk score you can read

A small demonstration of how routine patient data can flag who may be at higher risk of returning to hospital within 30 days, using a points score instead of a black box.

By Yashi Srivastava Lucknow, Uttar Pradesh June 2026 Python · pandas · scikit-learn
Educational demonstration only. This project uses a synthetic dataset I generated myself. The score is not a clinical prediction tool and must not be used to make decisions about any real patient.
Project summary

I built a simple, interpretable risk score for 30-day hospital readmission using a synthetic dataset of 80 patients. After cleaning the data (duplicates, a typo in age, missing adherence values), I fitted a logistic regression on four yes/no risk flags and converted it into a 0 to 13 point scorecard that anyone can add up by hand. Patients in the high-risk tier were readmitted at about four times the rate of the low-risk tier, and that tier held 62% of all readmissions while covering 36% of patients. The two strongest signals were a chronic condition and medication adherence below 70%.

Dashboard snapshot

High-risk patients were readmitted about 4x as often as low-risk patients

Patients per risk tier, split by whether they were readmitted within 30 days (n = 80).

Readmitted Not readmitted

Source: synthetic Sehat Watch dataset, 80 patients after cleaning.

20%
overall 30-day readmission rate (16 of 80)
62%
of readmissions fall in the high tier, which holds 36% of patients
0.74
AUC of the points score (0.67 under 5-fold cross-validation)

The risk score

Each flag is worth points in proportion to its logistic regression coefficient, scaled so the weakest flag counts 1 point. Add up the flags that apply.

Scorecard
Risk flagOdds ratioPoints
Medication adherence below 70%2.505
Has a chronic condition2.495
Age 65 or older1.522
3 or more visits in past 12 months1.191
Low: 0 to 2 Medium: 3 to 5 High: 6 to 13
Try the scorer
0Low risk

Dataset

80 synthetic patients, generated in Python with a fixed random seed so the results can be reproduced. Variables: age, gender, visits in the previous 12 months, a chronic-condition indicator (1 = yes), medication adherence (% of prescribed doses taken) and the outcome, readmission within 30 days (1 = yes). Adherence values shown in blue were missing and filled with the median.

IDAgeSexVisits 12mChronicAdherence %ReadmittedPointsTier

Data preparation

  1. Removed 2 duplicate records (same patient ID entered twice), bringing 82 raw rows down to 80.
  2. Fixed 1 implausible age: 470 was clearly a typing error for 47.
  3. Filled 6 missing adherence values with the median (78%) and kept a flag column so imputed rows stay traceable.
  4. Turned each variable into a yes/no flag with a readable cut-off: age 65+, 3+ visits, chronic condition, adherence below 70%.

Findings

  1. Adherence and chronic illness drive most of the risk. Each roughly 2.5x the odds of readmission. Patients with adherence below 70% were readmitted 38% of the time, against 12% for everyone else.
  2. A short list catches most readmissions. The high tier (6+ points) covered 29 of 80 patients but 10 of the 16 readmissions, a 34% rate against 8% in the low tier.
  3. Visit count adds little once chronic status is known. Frequent visitors were mostly chronic patients already, so the visits flag only earned 1 point after adjustment.

Healthcare and business implications

Target adherence support first

Pharmacist calls, refill reminders or blister packs aimed at low-adherence chronic patients address the two strongest flags at once, and adherence is the one flag a care team can actually change.

Use a ranked list, not a yes/no

A hospital with limited follow-up staff can call the high-tier list first. Here that means contacting about a third of discharges to reach most of the likely readmissions.

Methodology

I generated the dataset in Python with NumPy, building in realistic relationships (older age, chronic illness, more visits and lower adherence all raise risk) plus some noise and data-entry problems to clean. After preparation in pandas, I fitted a logistic regression in scikit-learn on the four flags. I divided each coefficient by the smallest one and rounded to whole numbers to get the points, then grouped totals into three tiers. I measured discrimination with AUC, both on the full dataset and with 5-fold cross-validation, and checked observed readmission rates within each tier.

Model limitations

  • The data is synthetic, so the relationships reflect my assumptions, not real patients.
  • 80 records is very small. Coefficients are unstable and the cross-validated AUC (0.67) is noticeably lower than in-sample (0.74), a sign of overfitting.
  • There was no separate test set, and the tier cut-offs were chosen on the same data.
  • Turning variables into yes/no flags loses detail (a 64-year-old and a 30-year-old score the same).
  • Key real-world drivers such as diagnosis, length of stay, lab results, social support and income are missing.
  • The score has not been checked for fairness across gender or other groups.

Suggested dashboard screenshot

For the portfolio card, capture the Dashboard snapshot section above: the tier chart on the left and the three figures on the right (20% overall readmission, 62% of readmissions in the high tier, AUC 0.74). Crop it so the scorecard table sits just underneath, which shows the score is readable by hand. Use "Readmission risk you can read" as the caption and keep the educational-demo note visible in the card text.

Disclaimer

This is an educational demonstration built on synthetic data to show how interpretable risk scoring works. It is not a clinical prediction tool, has not been validated on real patients, and should not be used for diagnosis, treatment or any decision about patient care.
Sehat Watch · Predictive Healthcare Analytics · Yashi Srivastava · Lucknow · June 2026

PROJECT #10

PROJECT #10

HealthTech Competitive Landscape
Sehat Watch · Market Intelligence Project

HealthTech Competitive Landscape: remote monitoring and connected care

Nine companies that monitor patients outside a doctor's line of sight, compared on who they serve, what they sell, how they use data and where they compete. The map shows a clear gap: clinical-grade monitoring at home is almost empty in India.

By Yashi Srivastava Lucknow, Uttar Pradesh August 2026 Category: Remote monitoring and connected care
Project summary

I mapped nine companies in remote monitoring and connected care, seven Indian and two global benchmarks, across target market, customer, product, AI capability, business model, geography and differentiation. Placed on a 2x2 of care setting against clinical acuity, the Indian players split into two camps: hospital and ICU monitoring (Dozee, Cloudphysician, Tricog) and consumer diabetes and wellness (BeatO, Sugar.fit, Fitterfly, Ultrahuman). The quadrant between them, clinical-grade monitoring at home after discharge, is held only by US-focused Biofourmis. That gap, and what it would take to fill it, is the main finding.

Competitive positioning map

Indian players cluster at the two ends; clinical home monitoring is open

Horizontal: where care happens and who buys. Vertical: how clinical the decisions are. Positions are my judgement from public sources.

CLINICAL CARE AT HOME white space in India HOSPITAL WELLNESS (EMPTY) CONSUMER WELLNESS AND COACHING HOSPITAL CLINICAL Home, consumer pays Hospital or ICU, provider pays Care setting and buyer Clinical acuity Lifestyle Acute decisions Ultrahuman: consumer wearable, wellness Ultrahuman BeatO: home diabetes monitoring and care BeatO Sugar.fit: CGM-led diabetes program Sugar.fit Fitterfly: digital therapeutics, employer channel Fitterfly Biofourmis: hospital at home and RPM for US health systems Biofourmis Tricog: AI ECG and virtual cardiology Tricog Dozee: contactless ward monitoring and early warning Dozee Philips: hospital patient monitoring and eICU Philips Cloudphysician: AI-assisted tele-ICU Cloudphysician
Indian company Global benchmark

Hospital clinical

Dozee, Cloudphysician, Tricog, Philips. Sold to hospitals on outcomes and staff leverage. Most mature and most evidence-backed.

Consumer wellness and coaching

Ultrahuman, BeatO, Sugar.fit, Fitterfly. Crowded, diabetes-heavy, paid by consumers or employers.

Clinical care at home

Only Biofourmis, and only in the US where Medicare pays for remote monitoring. No Indian player is built here yet.

Hospital wellness

Empty by design. Hospitals do not buy lifestyle tools.

Competitor comparison

CompanyTarget marketCustomer typeProduct categoryAI / data capabilityBusiness modelGeographic focusDifferentiation
Hospital and clinical monitoring
DozeeBengaluru General wards and step-down beds that lack continuous monitors Hospitals, nursing facilities; some home use Contactless vitals sensor (under-mattress) plus early warning system Ballistocardiography signal processing; AI early warning, company claims up to 16 hours ahead Device plus monitoring platform sold to hospitals India first; also US and Africa (300+ hospitals, 4 countries) No wires on the patient; US FDA 510(k) clearance
CloudphysicianBengaluru ICUs in hospitals without round-the-clock intensivists Hospitals (ICU and emergency) Tele-ICU service on its RADAR care centre platform AI-augmented monitoring and clinical co-pilot for remote intensivists Full-stack virtual critical care service (software plus specialists) 200+ hospitals in 23 Indian states; plans for emerging markets and US Sells specialist time, not just software; founded by intensivists; reports up to 40% lower ICU mortality
Tricog HealthBengaluru Cardiac diagnosis where cardiologists are scarce Clinics, hospitals, diagnostic centres, government networks Cloud ECG (InstaECG) and remote echo (InstaEcho) AI ECG interpretation with cardiologist review; 3M+ patients screened Per-test virtual cardiology service India plus 12+ countries in Southeast Asia and Africa 2,500+ sites; fast heart attack triage at primary care level
PhilipsGlobal benchmark Hospital patient monitoring across acuity levels Hospitals and health systems Bedside monitors (IntelliVue) and eICU / tele-ICU software Clinical decision support on large installed monitor base Capital equipment plus software and service contracts Global; India tele-ICU deployments since 2013 Installed base, brand trust and full hospital portfolio
Home clinical monitoring
BiofourmisGlobal benchmark Post-discharge and hospital-level care at home Health systems (for example Lee Health, Florida) RPM and hospital-at-home platform with care teams FDA-cleared predictive analytics for deterioration; Epic EHR integration Tech-enabled care service paid by health systems United States Only player here at hospital-grade acuity at home; client reported 50% fewer 30-day readmissions
Consumer chronic care and wellness
BeatOGurugram People with diabetes, especially first-time testers Consumers Connected glucometer, app, coaching and doctor consults Glucose logs feed care plans and coaching Device plus care-plan subscriptions and supplies India; 70%+ users from Tier 2 cities and beyond Scale and affordability: 15 lakh+ members, 55% never used a glucometer before
Sugar.fitBengaluru Type 2 diabetes and prediabetes Consumers CGM-led program with coaching, diagnostics and specialists CGM data plus fitness tracking drives personalised plans Paid subscription program India, urban Premium CGM experience; reports 2% average HbA1c drop in 3 months
FitterflyMumbai Diabetes, obesity, PCOS and other metabolic conditions Employers and consumers Digital therapeutics programs with tracking and coaching Automated tracking and personalised diet, sleep and stress plans Program fees, including corporate wellness contracts India Employer channel with published outcomes (1.1 point HbA1c drop in 88% of a pilot)
UltrahumanBengaluru Health-conscious consumers tracking sleep, activity and metabolism Consumers Smart ring, glucose wearable (M1), blood testing (Blood Vision) Continuous sensor data turned into health scores and insights Hardware sales (91% of revenue) plus subscriptions Global; US 61% of FY25 sales, India under 3% Only profitable player here: Rs 565 cr revenue, Rs 73 cr profit in FY25

Scroll sideways to see all columns. Company names link to the main source used.

Strategic findings

  1. The home clinical quadrant is empty in India because nobody pays for it yet. Biofourmis works in the US because Medicare reimburses remote monitoring. In India, hospitals pay for in-hospital tools and consumers pay for wellness, but no payer covers post-discharge monitoring. The gap is a reimbursement gap more than a technology gap.
  2. Diabetes is the crowded default for consumer players. BeatO, Sugar.fit and Fitterfly all sell device-plus-coaching for diabetes. They now compete on channel and price point rather than technology: BeatO goes mass market in smaller cities, Sugar.fit goes premium with CGMs, Fitterfly sells through employers.
  3. Hospital-side winners sell outcomes, and Indian firms are building for export. Cloudphysician leads with ICU mortality, Dozee with nursing time and early warning. Meanwhile Ultrahuman earns 61% of revenue in the US, Tricog runs in 12+ countries and Dozee holds US FDA clearance. India works as a low-cost build and validation base for global sales.

Business implications

For a new entrant

The open opportunity is post-discharge monitoring sold to hospitals, not to patients. A hospital that pays for 30 days of home monitoring for cardiac or diabetic patients can cut readmissions and keep the patient in its network. Hospital-side players such as Dozee are best placed to stretch into this space.

For consumer diabetes platforms

Differentiation on features is fading. The next advantage is clinical evidence and payer access: published HbA1c results, insurer or employer contracts, and ABHA-linked records that let a doctor act on home data. Firms that only sell devices and coaching risk a price war.

Methodology

Desk research carried out in August 2026. I defined the category as companies that monitor patients outside a direct clinical encounter, then picked nine: seven Indian companies covering hospital, clinical and consumer segments, and two global benchmarks (Philips and Biofourmis) to show what mature models look like. For each I collected facts from funding announcements, company disclosures, financial filings reported in the press and partner case studies, and filled the seven comparison fields. Positions on the 2x2 are my own scoring on two axes, care setting and buyer (home to hospital) and clinical acuity (lifestyle to acute decisions), based on each company's main product.

Limitations

  • Nine companies is a sample, not a full market scan. Players such as Qure.ai, HealthifyMe and GE HealthCare were left out to keep the category tight.
  • Most figures are company-reported (hospital counts, outcome claims) and not independently verified.
  • Some sources are funding announcements from 2020 to 2024, so scale numbers are likely higher today.
  • 2x2 positions are qualitative judgements; another analyst could reasonably place some companies differently.
  • Private company revenues were not available except for Ultrahuman, so market share could not be compared.

Suggested project screenshot

Capture the positioning map with its four quadrant cards beside it. The shaded "clinical care at home" quadrant with only one grey dot carries the main finding at a glance. Use "Where India's remote monitoring market is, and where it isn't" as the card caption, and link the card to this page.

Sources

  1. Dozee company profile (hospitals, beds, countries, product)
  2. MobiHealthNews: Dozee Series A2 and US FDA clearance
  3. Outlook Business: Cloudphysician raises $10.5M led by Peak XV (June 2024)
  4. India Med Today: Tricog Series B and product reach
  5. IT Voice: Philips tele-ICU in India
  6. HLTH: Biofourmis and Lee Health hospital at home (Nov 2024)
  7. Entrepreneur India: BeatO raises $33M Series B
  8. BioSpectrum: Sugar.fit secures $11M (Nov 2023)
  9. The Week: Fitterfly corporate health results (Jan 2024)
  10. Entrackr: Ultrahuman FY25 revenue and profit
  11. Physicians Practice: 2026 Medicare RPM codes