A proposal from OHC Foundation to IndiaAI Mission Keralam
Building Healthcare Models from Keralam, for the World
From healthcare infrastructure to healthcare intelligence.
CARE · OPEN MODELS · OPEN BENCHMARKS · OPEN HEALTHCARE AI
The inevitable shift
Identity got rails. Payments got rails. Healthcare is next.
India built national rails for identity and payments. Care delivery is still fragmented across apps, schemes, hospitals, registers and vendors, with no shared operating layer underneath.
Past
Aadhaar
National identity rails
Present
UPI
National payment rails
Next
CARE
Public healthcare rails: patients, beds, visits, referrals, medicines, intelligence
IndiaAI is building the compute and the foundation models. Healthcare intelligence still needs its own rails.
OHC Foundation · IndiaAI Mission Keralam
02 / 33
The next transformation in healthcare
Healthcare is moving from software that records care to systems that understand care.
Last decade
Capture
→
Store
→
Retrieve
→
Report
Next decade
Listen
→
Understand
→
Structure
→
Reason
→
Assist
Healthcare AI will become part of the everyday workflow of every doctor, nurse, community health worker and health system.
The intelligence powering this future cannot be built only around English-language medicine and Western health systems. It has to understand how we speak, document and deliver care.
OHC Foundation · IndiaAI Mission Keralam
03 / 33
The problem
The world has powerful AI models. Very few understand Indian healthcare.
A consultation in Kerala
Malayalam + English + medical terminology + abbreviations + local drug names + incomplete sentences + clinical context.
A diagnostic report
Handwritten or printed, in thousands of formats from thousands of laboratories.
A discharge record
Years of history compressed into a few pages.
Beyond fluent text, healthcare AI must understand
- Symptoms and diagnoses
- Medications, dose, frequency
- Clinical negation
- Investigations and units
- Temporal history
- Clinical terminology
- Care pathways
- Referrals and follow-up
- Uncertainty
Healthcare needs models grounded in clinical reality, not just language fluency.
OHC Foundation · IndiaAI Mission Keralam
04 / 33
Why Open Healthcare Network · We start with healthcare, not with a model
CARE is already the open operating layer for healthcare delivery.
A reusable backbone for the workflows that usually become disconnected projects. Self-hostable, modular, standards-based, MIT-licensed, and a UN-recognised Digital Public Good since 2023.
One open layer beneath every care pathway.
Hospital workflows
Home and palliative care
ICU and TeleICU
Specialty care
Pharmacy and inventory
Referrals and follow-up
Public-health monitoring
AI-assisted documentation
That gives OHC something unusual for healthcare AI: a path from model → workflow → clinician → real-world evaluation.
OHC Foundation · IndiaAI Mission Keralam
05 / 33
Proof · Where CARE began
CARE started in Kerala's COVID war rooms in 2020.
A community project for real-time capacity monitoring and care coordination in Ernakulam, run with the district administration through all three pandemic waves.
465+
Hospitals onboarded
355K+
Patients managed
200K+
Telemedicine calls
35×
Patient surge in 40 days
The New York Times: “As India stumbles, one state charts its own Covid course.”
OHC Foundation · IndiaAI Mission Keralam · March 2020 to May 2022
06 / 33
Proof · Kerala Care, a Government of Kerala initiative
Kerala's state-wide home and palliative care grid runs on CARE.
Deployed April 2025 on the first version of the CARE Core architecture
1,412+
Government facilities
1,305+
NGOs in the network
374K+
Home-bound patients
1.9M+
Home-care consultations
12,000+
Government staff as users
100%
Paperless since Jan 2026
Nurses and doctors visiting homes, out-patient and in-patient facilities, comfort devices, volunteers and care homes, on one register. Every report is online. dashboard.care.kerala.gov.in
OHC Foundation · IndiaAI Mission Keralam
07 / 33
Proof · Across India
One open core now runs hospitals, ICUs and home care in 11 states.
216
ICUs connected across 15 TeleICU hubs
160M
People covered by CARE workflows across 7 states
11
States that deployed CARE in five years
100+
Hospitals live on CARE HMIS
UN Digital Public Good since 2023
MIT licensed
1,379+ contributors
No rip-and-replace
CARE HMIS went live in January 2026 and is scaling across Kerala, Karnataka, Jharkhand, Assam and Manipur. TeleICU powers the 10BedICU programme; oncology runs with the National Cancer Grid.
OHC Foundation · IndiaAI Mission Keralam
08 / 33
Why it compounds
Open standards make every deployment reusable.
FHIR-native architecture turns one state's deployment into infrastructure the next state, hospital or country can reuse. CARE 3.0 models 63 resources across 7 operational domains.
Connects
Hospitals · Public-health programs · ABDM · Devices
standards bridge
Feeds
Analytics · Public health · Research · AI models
FHIR R5
SNOMED CT
LOINC
UCUM
ICD-10
DICOM-ready
ABDM-ready
Terminology server
Open APIs
RBAC + audit
Every CARE deployment produces structured, coded, longitudinal records: the substrate healthcare models need.
OHC Foundation · IndiaAI Mission Keralam
09 / 33
Already shipping
AI already runs inside CARE workflows.
Voice
Ambient clinical documentation
Malayalam and English consultations become structured encounters.
Vision
Clinical document intelligence
Lab reports and clinical documents become structured observations.
Language
Clinical summarisation
Longitudinal records become discharge summaries, handovers and care plans.
Also in production or pilot: Care Scribe, AI record summaries, doctor copilot, document understanding, vision-assisted registration, discharge summary builder, care-plan generation and drug-safety alerts.
Today these are products. Tomorrow they become reusable healthcare models.
OHC Foundation · IndiaAI Mission Keralam
10 / 33
01
Clinical speech
02
Clinical documents
03
Clinical language
01 · SPEECH
00:14 · Malayalam + English
Consultation audio, captured in the room while the doctor talks to the patient.
→
02 · TRANSCRIPT
കഴിഞ്ഞ മൂന്ന് ദിവസമായി ചുമയുണ്ട്. Fever ഇല്ല.
“Cough for the last three days. No fever.”
Code-switched speech, kept as spoken.
→
03 · STRUCTURED DRAFT
Symptom
Cough
Duration
3 days
Fever
Absent
Clinician review required
Speech → Transcript → Structure → Review · the clinician approves before anything is written to the record
OHC Foundation · IndiaAI Mission Keralam · Care Scribe, ambient documentation in CARE
11 / 33
01
Clinical speech
02
Clinical documents
03
Clinical language
01 · DOCUMENT
LABORATORY REPORT
Haemoglobin
12.8 g/dL
WBC
7.1 ×10³/µL
Platelets
245 ×10³/µL
A printed report, photographed on a phone or uploaded as a PDF.
→
02 · EXTRACTED FIELDS
Test
Haemoglobin
Value
12.8
Unit
g/dL
Source
page 1 · row 1
Value, unit and source location kept together.
→
03 · OBSERVATION DRAFT
Observation
Haemoglobin
Value
12.8
Unit
g/dL
Verify against source
Document → Extract → Structure → Review · a coded observation lands in the record only after verification
OHC Foundation · IndiaAI Mission Keralam · Document understanding in CARE
12 / 33
01
Clinical speech
02
Clinical documents
03
Clinical language
01 · RECORDS
[1] DAY 1
Cough documented for three days.
[2] DAY 2
Cough improved. No fever recorded.
[3] NOTE
Review planned in one week.
Encounters, observations and notes from the structured record.
→
02 · CONTEXT
Cough · 3 days [1]
Improved · day 2 [2]
No fever [2]
Review · 1 week [3]
The clinically relevant facts, each tied to the record it came from.
→
03 · SUMMARY DRAFT
Three-day history of cough [1], improved on day two with no fever recorded [2]. Review planned in one week [3].
Review before finalising
Records → Context → Summary → Review · every statement stays connected to its source
OHC Foundation · IndiaAI Mission Keralam · Clinical summarisation in CARE
13 / 33
Our ambition
Move from using AI models to building healthcare models.
A family of open healthcare models, trained and evaluated for the realities of care.
Speech
Understand clinical conversations
Vision
Understand healthcare documents and images
Language
Understand and generate clinical records
Terminology
Connect natural language to structured medicine
Multimodal
Reason across voice, text, documents and clinical context
Agents
Use models safely inside clinical workflows
Small, specialised, deployable models where possible. Frontier models where necessary.
OHC Foundation · IndiaAI Mission Keralam
14 / 33
Start with Malayalam
Malayalam is our starting point, not our boundary.
Kerala gives us a powerful environment to solve one of the hardest healthcare AI problems: clinical language as it is actually spoken.
Not textbook Malayalam. Not textbook English. Malayalam + English + medical vocabulary + local expressions + clinical shorthand.
A real consultation sounds like
“Sugar control കുറച്ച് മോശമാണ്… metformin continue ചെയ്യാം… creatinine വീണ്ടും check ചെയ്യണം.”
If we can build models that understand this reliably, the same architecture can extend across India's languages and, eventually, other multilingual health systems.
Malayalam
→
Indian languages
→
Global South
OHC Foundation · IndiaAI Mission Keralam
15 / 33
Models are only half the problem
You cannot improve what you cannot measure.
India needs healthcare benchmarks that answer one question: does this model actually understand our patients and clinicians? Not just “what is the word error rate?” but “did it preserve the clinical meaning?”
Health Bench
A multilingual benchmark for healthcare AI. Start with Malayalam.
Open evaluations, open test suites, open results: a shared yardstick for everyone building healthcare AI in India.
OHC Foundation · IndiaAI Mission Keralam
16 / 33
What Care Health Bench measures
Benchmark healthcare AI the way clinicians experience failure.
Clinical Speech
- Malayalam
- English
- Code-switching
- Accents
- Medical vocabulary
Clinical Understanding
- Symptoms
- Diagnoses
- Medications
- Allergies
- Procedures
- Temporality
Clinical Safety
- Negation
- Medication dosage
- Omitted information
- Unsupported information
- Critical values
Clinical Documents
- Lab extraction
- Units
- Reference ranges
- Document layouts
- Noisy captures
Clinical Generation
- Discharge summaries
- Encounter summaries
- Referrals
- Care plans
- Longitudinal synthesis
The benchmark becomes common infrastructure for everyone building healthcare AI in India.
OHC Foundation · IndiaAI Mission Keralam
17 / 33
From benchmark to commons
We want to build more than models. We want to build a commons.
CARE deployments
→
Real clinical workflows
→
Governed, de-identified datasets
→
Open benchmarks
→
Healthcare models
→
Real-world clinical evaluation
→
Human feedback
↺ every deployment compounds the commons
At the centre
Health AI Commons
Models · Datasets · Benchmarks · Evaluation tools · Terminology · Deployment recipes
The commons holds
- Clinical workflow and care-pathway datasets
- Multilingual medical language and voice data
- Evaluation benchmarks and safety test suites
- FHIR-native training and inference pipelines
- Human feedback from doctors, nurses and program teams
OHC Foundation · IndiaAI Mission Keralam
18 / 33
Build once. Improve everywhere.
Every healthcare deployment can make the intelligence layer better.
Traditional healthcare software
Every hospital becomes another implementation.
Our model
Every implementation can become another learning environment.
Not by centralising patient data, but through:
Common standards
Governed datasets
Shared evaluation
Federated learning where appropriate
Reusable model improvements
Common safety infrastructure
The network compounds.
OHC Foundation · IndiaAI Mission Keralam
19 / 33
Open healthcare intelligence
Healthcare intelligence should become infrastructure, not another vendor lock-in.
Models
Healthcare-specific weights and adapters where licensing and data governance permit
Benchmarks
Multilingual clinical AI evaluations
Datasets
Governed and appropriately de-identified research datasets
Evaluation
Safety tests and clinical fidelity metrics
Standards
FHIR-native model inputs and outputs
Tools
Training, inference and deployment pipelines
Build together. Evaluate together. Improve together.
OHC Foundation · IndiaAI Mission Keralam
20 / 33
Why Keralam
Kerala has the best health outcomes in India.
Healthcare AI needs three things at population scale: outcomes worth learning from, clinicians who document care, and delivery systems that reach every home. Kerala has all three. The best healthcare AI should be built where the best healthcare is.
#1
NITI Aayog State Health Index, large states
Four rounds in a row
75.1
years of life expectancy at birth
Highest in India · SRS 2019–23
5
infant deaths per 1,000 live births
India: 25 · SRS 2023
20
maternal deaths per lakh live births
India: 93 · SRS 2019–21
Near-universal access to palliative care
Asia's first palliative care policy, 2008
Trained home-visit teams in ~900 panchayats
Aardram mission
Kerala Care grid already on CARE
Sources: SRS, Registrar General of India · NITI Aayog State Health Index 2019–20 · Govt of Kerala DHS
21 / 33
Why Keralam
Keralam can become a global laboratory for responsible healthcare AI.
Healthcare
The best outcomes in India, real clinical environments and public institutions
AI
IndiaAI compute, researchers and the Indian model ecosystem
Infrastructure
CARE and open healthcare workflows already running state-wide
Language
Malayalam as a high-value multilingual AI frontier
Kerala already gives us the Kerala Care grid, CARE HMIS hospitals, clinician partners and Malayalam clinical speech from real consultations.
Bring researchers, clinicians, models and real healthcare workflows into one ecosystem.
Build here. Validate here. Open it to the world.
OHC Foundation · IndiaAI Mission Keralam
22 / 33
From Keralam to the world
The problems Kerala can solve are not uniquely Kerala's problems.
Kerala
- Malayalam
- Code-switching
- Public healthcare
- Community care
- Hospitals
- Resource constraints
→
India
- 22+ languages
- Massive clinical diversity
- Public + private healthcare
- Rural + urban systems
→
Global South
- Multilingual care
- Workforce shortages
- Fragmented infrastructure
- Limited specialist access
A healthcare model that works only in a Silicon Valley clinic is a product. A model that adapts across languages, systems and resource settings can become infrastructure.
OHC Foundation · IndiaAI Mission Keralam
23 / 33
What we want to build with IndiaAI Mission Keralam
A Healthcare AI Lab inside the Open Healthcare Network ecosystem.
01
Healthcare Models
Speech · Vision · Language · Multimodal
02
Care Health Bench
Malayalam-first healthcare AI benchmark
03
Healthcare Datasets
Governed datasets for Indian clinical AI research
04
Clinical Evaluation
Real-world evaluation with clinicians
05
Open Model Infrastructure
Training · Evaluation · Inference · Deployment
06
Research Community
Clinicians + engineers + universities + startups + government
Create the infrastructure that lets many healthcare AI teams build faster.
OHC Foundation · IndiaAI Mission Keralam
24 / 33
The first models
Start with problems where AI can remove friction from care today.
Clinical Speech Model
- Malayalam–English clinical ASR
- Ambient conversations
- Medical terminology
- Speaker-aware transcription
- Structured encounter generation
Clinical Document Model
- Lab reports
- Prescriptions
- Discharge records
- Forms
- Clinical document understanding
Clinical Language Model
- Summarisation
- Clinical information extraction
- Care-plan generation
- Terminology mapping
- Structured clinical outputs
Later
Multimodal reasoning · Imaging · Decision support · Patient communication · Clinical agents
OHC Foundation · IndiaAI Mission Keralam
25 / 33
The model architecture
Not one giant model. A healthcare model ecosystem.
Foundation models
↓ specialised using
Indian healthcare data · Medical terminology · CARE workflows
↓
Speech model
Vision model
Clinical language model
Multimodal model
↓
Common layer: Care Health Bench · Safety · Terminology · FHIR
↓
Scribe
Lab digitisation
Summaries
Coding
Decision support
Agents
OHC Foundation · IndiaAI Mission Keralam
26 / 33
Intelligence embedded in care
The goal is intelligence inside care, not an AI app beside it.
AI
inside
Patient registration
Consultation
Orders
Diagnostics
Medication
Referral
Discharge
Home care
Follow-up
It listens when necessary. Reads when necessary. Structures automatically. Surfaces intelligence at the right moment.
The most valuable healthcare AI will disappear into the workflow.
OHC Foundation · IndiaAI Mission Keralam
27 / 33
Human-centric by design
AI drafts. Humans decide. CARE records.
Every model deployed through OHC supports:
Human review
Source traceability
Structured outputs
Role-based access
Auditability
Model and version provenance
Rollback
Continuous evaluation
In healthcare, trust is not a feature of the model. It is a property of the system around the model.
Governed by consent, audit logs, role-based access, output traceability, clinical review and rollback, aligned with DPDP. AI a clinician can trust, and a regulator can verify.
OHC Foundation · IndiaAI Mission Keralam
28 / 33
Who we are
A foundation built to steward open healthcare infrastructure.
Srikanth Nadhamuni
Founding CTO, Aadhaar · Chairman, 10BedICU
Sandeep Singal
Board Member, ACT Grants
Dr. Athul Joseph Manuel
Chief Medical Officer · President, IMA Kochi
Bodhish Thomas
CEO · Product, AI, foundation leadership
Vignesh Hari
CTO · Architecture, product and engineering
Sanjay Vijayakumar
Executive Director · Strategic operations
Gigin Chandy George
CTO, eGov Healthcare
Pinaki Krishna
Resources and partnerships
1,379+ contributors to the code base and 6,000+ community members in Slack
Open Healthcare Network Foundation, a Section 8 company since December 2025, is the steward of CARE. Backed by ACT, eGov Foundation, Schmidt Futures, MSDF, Facebook and Balagopal Chandrasekhar.
OHC Foundation · IndiaAI Mission Keralam
29 / 33
Community signals
Recognition from people who understand infrastructure.
Satya Nadella
CEO, Microsoft
“I had a chance to meet with the Open Healthcare Network, it's fantastic to see what they are doing.”
Kyle Daigle
COO, GitHub
“The Open Healthcare Network in India is a powerful example of people coming together to advance human progress, solving the problems that matter.”
Kai Chen
OpenAI
“One of the coolest non-profits operating in the healthcare space… they bring advanced ICU care to places in rural India.”
Sreevas Sahasranam
University of Glasgow
“From a COVID dashboard fully run by volunteers for a district in Kerala to this… almost like a unicorn valuation equivalent for the social sector.”
Srikanth Nadhamuni
Founding CTO, Aadhaar
“One of the most active open source contributor communities in India.”
OHC Foundation · IndiaAI Mission Keralam
30 / 33
The opportunity
India can become a producer of healthcare intelligence, not just a consumer of it.
The bet
India should build intelligence from the realities of its own healthcare system, and make that intelligence useful to the world.
IndiaAI
gives us compute.
Kerala
gives us healthcare, language and institutions.
CARE
gives us the operating layer.
OHC
brings the open-source ecosystem.
Together, we can build the missing healthcare AI layer.
OHC Foundation · IndiaAI Mission Keralam
31 / 33
The long-term vision
The Linux of Healthcare Intelligence
An open ecosystem where governments, hospitals, researchers and companies can train, evaluate, deploy and improve healthcare AI without rebuilding the foundation every time.
Healthcare infrastructure below.
Healthcare intelligence above.
An open commons between them.
From Keralam.
For the world.
Build the world's open healthcare intelligence layer with us.
Contact
Bodhish Thomas
CEO, Open Healthcare Network Foundation · bodhish@ohc.network
Download this deck
pitch.ohc.network
ohc.network · github.com/ohcnetwork
CARE · MODELS · BENCHMARKS · COMMONS