Every hospital owner we speak to has the same quiet worry. The AI demo looked brilliant in the meeting room. But what happens when it gets a patient's report wrong, when an auditor asks how the model was tested, or when a state health department asks whether it works on Indian patients at all?
For a long time, nobody in India could answer those questions with a clear public reference point. That is now changing. In the first week of November 2026 the National Health Authority (NHA) moved closer to a formal implementation roadmap for India's AI-in-healthcare strategy, and it showed an AI-assisted sandbox that completed an ABDM integration step live, in front of state officials.
If you are a hospital, a clinic chain, a diagnostic centre or a health-tech founder planning an AI-enabled app, this guide explains what changed, what it means for your product, and how to build in a way that will still look sensible when the rules get firmer.
Quick answer: India now has a national strategy for AI in healthcare (SAHI, launched February 2026) and a public benchmarking platform (BODH) for testing health-AI models on Indian data. NHA is finalising a roadmap that assigns SAHI's recommendations to agencies, and its Sandbox 2.0 uses agentic AI to speed up ABDM integration. Build your app with human sign-off on anything clinical, evidence of testing on Indian data, consent-first data flows, and ABDM-ready architecture from day one.
What actually happened this month
Three public developments matter for anyone building health software in India.
1. SAHI is moving from strategy to an action plan
The Strategy for Artificial Intelligence in Healthcare for India (SAHI) was launched by Union Health Minister J.P. Nadda on 17 February 2026 at the India AI Impact Summit, according to the Press Information Bureau. The government described it as a governance framework and national roadmap for ethical, transparent and accountable use of AI in healthcare.
On 2 November 2026, NHA CEO Dr Sunil Kumar Barnwal chaired the fourth committee meeting on the SAHI Implementation Roadmap, reported by Elets eHealth. The roadmap assigns SAHI's 32 recommendations to lead and supporting agencies with timelines and resource needs. No release date has been announced yet.
SAHI's five pillars are:
- Governance, regulation and trust
- Health data and digital infrastructure
- Workforce, institutional capacity and change management
- Research, innovation and evidence generation
- Ecosystem enablement and global leadership
2. BODH gives health AI a public test bench
BODH (Benchmarking Open Data Platform for Health AI) was launched alongside SAHI. It was developed by IIT Kanpur with NHA, and PIB says it is designed to assess the performance, robustness, bias and generalisability of AI systems using diverse, anonymised real-world health datasets before deployment at scale.
At NHA's national review meeting in Kolkata on 5–6 November 2026, BODH was presented as a way for states to evaluate health-AI models on Indian datasets before procurement.
3. Sandbox 2.0 uses agentic AI for ABDM onboarding
At the same Kolkata meeting, NHA demonstrated Sandbox 2.0 enhanced with agentic AI. During the demo, an ABDM Milestone 1 integration for ABHA creation was completed in real time, according to Elets eHealth's report of the meeting, which cites the official PIB release. NHA also showed an Ambient AI Voice Scribe that turns doctor–patient conversations into structured records in multiple Indian languages. The doctor reviews and approves the draft before it is linked to the patient's ABHA.
Key takeaways
- SAHI (Feb 2026) is India's national framework for AI in healthcare; its implementation roadmap is being finalised.
- BODH lets models be tested for performance, bias and generalisability on anonymised Indian data.
- Sandbox 2.0 adds agentic AI to ABDM integration, which should shorten onboarding for HIS and app vendors.
- The direction is clear: evidence, human oversight and interoperability will be expected, not optional.
Why this matters even if you are "just" building an appointment app
Many teams think national AI policy only matters for radiology start-ups. It doesn't. The moment your app summarises a report, suggests a department, triages symptoms, drafts a discharge note or answers patient questions with a language model, you are building AI into a clinical journey.
Buyers are already asking about this. State programmes can now point to a public benchmarking platform. Hospital procurement teams will soon start asking the same questions in their tenders:
- What data was this model trained and tested on?
- Has it been evaluated on Indian patients, languages and devices?
- Who approves the output before it reaches a patient record?
- How do you monitor it after go-live?
- Can it plug into ABDM and our existing HIS?
If your product can't answer these, it will lose to one that can, even if your demo looks better.
The four design rules we follow for healthcare AI apps
Rule 1: Keep a clinician in the loop for anything clinical
India's Telemedicine Practice Guidelines say AI or machine-learning platforms cannot counsel patients or prescribe medicines; only a registered medical practitioner can. NHA's own voice scribe follows the same pattern: AI drafts, the doctor approves.
Design for that from the first wireframe:
- AI outputs are marked as drafts until a named clinician approves them
- Every approval is logged with time and user
- Patients never receive a diagnosis or prescription from a bot
- Administrative automation (booking, reminders, directions, document collection) is kept apart from clinical advice
Rule 2: Prove it works on Indian data
BODH signals what "good evidence" will look like: performance, robustness, bias and generalisability, measured on local data. Even before any formal requirement, build a simple validation pack:
- A description of training and test data sources, with consent basis
- Accuracy numbers on a held-out Indian test set, not just a global benchmark
- Results split by age group, sex, language and device type where relevant
- Known failure cases and what the app does when confidence is low
If your software is intended to support diagnosis or treatment decisions, also take regulatory advice on whether it counts as a medical device under India's Medical Devices Rules, 2017. That decision shapes your whole timeline, so make it early.
Rule 3: Consent and privacy first, not as a patch
The DPDP Rules, 2025 come into force in phases. The Government's announcement set the core duties (notices, consent, security safeguards, breach reporting) to apply from 13 May 2027, and the Consent Manager provisions start on 13 November 2026. For an AI app that means:
- Clear, specific consent for each purpose, including any use of data to improve models
- Data minimisation: don't send full records to a third-party model when a summary will do
- Hosting and vendor contracts that say where data goes and how long it stays
- A way for patients to withdraw consent and have their data handled accordingly
Rule 4: Build ABDM-ready architecture from day one
Sandbox 2.0 is meant to make ABDM integration faster. That only helps if your product is built to integrate. Plan for:
- ABHA creation and linking flows (Milestone 1) as a standard module
- Health record sharing with consent artefacts, using standard formats your HIS partners support
- Clean APIs so the AI layer sits beside the record system, not inside it
- Audit logs that a hospital's IT team can actually read
Comparison: a demo AI app vs a SAHI-ready AI app
- Clinical output. Typical "demo" AI app: Shown straight to patient. SAHI-ready healthcare AI app: Draft, approved by a named clinician.
- Evidence. Typical "demo" AI app: Global benchmark screenshot. SAHI-ready healthcare AI app: Validation on Indian data, split by key groups.
- Bias and failure cases. Typical "demo" AI app: Not documented. SAHI-ready healthcare AI app: Documented, with low-confidence fallback.
- Consent. Typical "demo" AI app: One checkbox for everything. SAHI-ready healthcare AI app: Purpose-specific consent, withdrawal supported.
- Interoperability. Typical "demo" AI app: Stand-alone database. SAHI-ready healthcare AI app: ABDM-ready APIs and HIS integration.
- Monitoring after launch. Typical "demo" AI app: None. SAHI-ready healthcare AI app: Error reporting, drift checks, periodic review.
- Procurement readiness. Typical "demo" AI app: Sales deck. SAHI-ready healthcare AI app: Validation pack, data-flow map, audit logs.
Where AI helps most right now (and where to be careful)
Lower-risk, high-value features you can ship first:
- Appointment and OPD flow: slot suggestions, reminders, queue updates, multilingual directions
- Documentation support: draft notes and discharge summaries for doctor approval
- Front-desk and call automation: answering timing, location and preparation questions; see our guide to AI voice agents for hospitals and clinics
- Search and summaries inside the patient portal: helping patients find their own reports, explained in plain language, with a "speak to your doctor" route
Higher-risk features that need validation, clinical governance and possibly regulatory review:
- Symptom triage that changes how urgently someone is seen
- Image or report interpretation
- Medication suggestions of any kind
- Risk scores used for admission or discharge decisions
The goal is not to avoid AI. It is to launch the safe, useful parts quickly while the evidence for the riskier parts is being built.
A practical 60-day build plan
Days 1–15: Scope and data map. List every AI feature and mark it administrative or clinical. Map where patient data enters, where it is stored, and which vendors touch it.
Days 16–30: Architecture and consent. Design the ABDM modules, consent screens and approval workflow. Pick model providers and hosting with clear data terms.
Days 31–45: Build and validate. Build the administrative features first. For clinical features, assemble an Indian test set and record accuracy, bias checks and failure cases.
Days 46–60: Pilot with oversight. Launch with one department. Track errors, overrides by doctors and patient complaints weekly. Write the validation pack your next buyer will ask for.
For the interface side, our article on building user-friendly medical portals covers patient-facing design, and building AI-powered features with LLMs explains the technical patterns.
How Pengwin Solutions builds healthcare AI apps
Pengwin Solutions is an AI-led app development and digital marketing team working with hospitals, clinics and health businesses. On healthcare products we:
- Separate administrative AI from clinical AI and design approval workflows for doctors
- Build ABDM-ready architecture and HIS integrations
- Set up consent, logging and monitoring before launch, not after an audit
- Help you prepare the validation and data-flow documents buyers and partners expect
- Market the finished product ethically through SEO and performance campaigns
See our app development services and healthcare services for more detail.
Frequently asked questions
What is SAHI in healthcare?
SAHI is the Strategy for Artificial Intelligence in Healthcare for India, launched by the Union Health Ministry in February 2026. It is a national framework with 32 recommendations across five pillars to guide safe, ethical and evidence-based use of AI in India's health system.
What is BODH and who can use it?
BODH is the Benchmarking Open Data Platform for Health AI, developed by IIT Kanpur with the National Health Authority. It is designed to evaluate health-AI models for performance, robustness, bias and generalisability on anonymised Indian datasets before large-scale deployment.
Is SAHI a law that my healthcare app must follow?
SAHI is a strategy, not a law. NHA is finalising an implementation roadmap that assigns its recommendations to agencies. Existing rules such as the Telemedicine Practice Guidelines, the DPDP Act and medical device regulations still apply today.
What is NHA Sandbox 2.0?
Sandbox 2.0 is NHA's ABDM integration environment enhanced with agentic AI. At the November 2026 Kolkata review it completed an ABDM Milestone 1 ABHA-creation integration live, which points to faster onboarding for app and HIS vendors.
Can an AI app give patients a diagnosis in India?
No. The Telemedicine Practice Guidelines say AI platforms cannot counsel patients or prescribe; a registered medical practitioner must make clinical decisions. AI can draft notes or flag information for a doctor to review.
How long does it take to build an AI-enabled healthcare app?
It depends on scope. A focused pilot with administrative AI features and ABDM modules can often be planned and piloted in about two months, while clinical AI features need extra time for validation and governance.
Plan your healthcare AI app with Pengwin
Call +91 90326 18332, email hello@pengwinsolutions.com, or contact Pengwin Solutions to discuss a healthcare app that is useful for patients, trusted by doctors and ready for India's AI rules.