Introducing the MedOS Gen AI Addon: 10 AI features built for Indian hospitals

Today we're launching the MedOS Gen AI Addon — 10 AI-powered features built into the workflows that Indian hospital staff use every day, powered by Google Gemini 2.5 Flash.

This isn't a chatbot. It's not a "summarise this for me" toy. The Gen AI Addon is a precision tool that lives where the work happens — inside the consultation page, the discharge dialog, the lab order, the NABH document creator, the insurance claim form, the analytics dashboard, and the incident investigation panel.

The 10 features

1. Clinical Assistant — During a consultation, click "AI Clinical Assist" and get top 3 differential diagnoses (ranked with reasoning), suggested investigations (cost-effective Indian-context choices), and an empirical treatment plan with NMC-compliant generic medicine names. Cites Indian guidelines (ICMR, MoHFW) where relevant.

2. Discharge Summary auto-draft — When you open the discharge dialog for an inpatient, click "AI Draft Summary" and get a full SOAP-format discharge summary auto-generated from the admission record, all linked consultations, abnormal lab results, and nursing notes. Review, edit, save.

3. Lab Report Explainer — On any released lab order, click "AI Explain Results" and get plain-English interpretation of abnormal values — one explanation for the doctor (clinical context, next steps) and another for the patient (what does this mean, what should I do).

4. NABH SOP Draft Generator — Creating a Quality Document for NABH compliance? Type the title and department, click "AI Draft SOP", and get a full 8-section SOP (Purpose, Scope, Definitions, Responsibilities, Procedure, Documentation, References, Revision History) ready to customise.

5. Insurance Claim Narrative — Drafting a TPA claim? Click "AI Draft Narrative" with the diagnosis, ICD-10 code, and procedures, and get a structured clinical justification ready to attach to the claim.

6. Analytics Narrator — On the analytics dashboard, click "AI Explain Revenue Trend" to get a plain-English explanation of what your revenue/OPD/operational charts are showing, with optimisation suggestions.

7. Consultation Template Auto-Generator — Creating a new consultation template? Describe what you need ("Routine T2DM follow-up: HbA1c review, foot exam, BP, lifestyle counseling") and the AI generates a complete structured template.

8. Emergency Triage Assistant — At ED quick registration, click "AI Suggest Triage" with chief complaints + vitals and get a triage level (RED/YELLOW/GREEN), reasoning, immediate action checklist, and red flags to watch.

9. Incident RCA Assistant — Investigating a medication error or fall? Click "AI Suggest RCA" and get fishbone-categorised likely root causes, a 5-whys chain, and CAPA suggestions.

10. Support Reply Suggester — On the support module, click "AI Suggest Reply" to draft a professional first response in MedOS support voice.

Why this matters

Indian hospitals are time-starved. The average doctor sees 40-60 OPD patients per day. The average front-desk staff handles 200+ touchpoints. The average billing person processes claims for 21+ TPAs with different formats.

Every feature in the Gen AI Addon was chosen because we asked clinic staff: "What do you spend the most time on that an AI could meaningfully accelerate?" The answers were: discharge summaries (45 mins each), claim narratives (20 mins each), NABH SOPs (3-4 hours each), lab report patient explanations (10 mins each).

The math: a 30-bed hospital doing 5 discharges per day saves about 3.5 hours of physician time daily. That's a real number.

How we handle privacy (the part that matters)

This is the part we obsessed over. Indian healthcare data is governed by DPDP Act 2023, DISHA, HIPAA-aligned safeguards, the HIV/AIDS Act 2017, the Mental Healthcare Act 2017, and the PCPNDT Act 1994. Sending raw patient data to a US-hosted LLM would be a non-starter.

So we don't.

Layer 1 — Field-level redaction. Before any data leaves MedOS, every known PHI field (patient name, phone, Aadhaar, ABHA, UHID, email, PAN, guardian/father/mother/spouse names, emergency contacts, policy numbers) is stripped and replaced with safe tokens like `[PATIENT_1]`, `[PHONE_1]`, `[AADHAAR_1]`.

Layer 2 — Regex sweep on free text. Consultation notes, complaints, examination findings — anywhere a patient might have typed an identifier — gets scanned with regex patterns for embedded phone numbers (Indian 10-digit format), Aadhaar (UIDAI 12-digit format with first digit 2-9), ABHA IDs, emails, PAN numbers, and multi-segment UHIDs. Any match gets tokenised too.

Layer 3 — Defensive assertion. Right before the API call to Gemini, we run the regex sweep one more time on the final payload. If any raw PHI is detected, the call is aborted with a security incident logged.

Layer 4 — Restoration server-side only. When Gemini's response comes back, we restore the safe tokens back to the original values, but only on the server, only for the duration of that request. The token-to-value mapping is never persisted.

Hard-blocked categories. HIV/AIDS records (ICD-10 B20-B24), Mental Healthcare Act records, and PCPNDT records are blocked entirely from AI processing — not redacted, refused. Individual patients can also opt out via an `ai_opt_out` flag.

Audit trail. Every AI call writes a row to `ai_usage_log` with clinic, user, feature, tokens, cost, latency, model, and success/error. 6-year retention. Audit logs also record AI_ADDON_ENABLED, AI_ADDON_DISABLED, and AI_SECURITY_INCIDENT events.

Per Google's policy: Gemini API data is not used to train AI models.

Cost & fair use

The Gen AI Addon is ₹699/month on Professional plans and included at no extra cost on Enterprise plans.

Fair-use caps: - Professional: 7,500 AI calls / month / clinic - Enterprise: 50,000 AI calls / month / clinic

Per-user rate limit: 20 calls/minute. Per-clinic rate limit: 200 calls/minute.

You can also disable specific AI features (e.g., turn off Insurance Claim Narrative if you don't process insurance) without disabling the whole addon.

Try it

Visit app.med-os.in/demo and you'll see the AI Powered badges and AI buttons across all 10 modules. AI calls themselves are blocked in demo mode (cost protection — anyone can click "Enter Demo" without signup), but you can see exactly where the AI lives in your workflow.

To actually run the AI features, sign up for a free 14-day Professional trial at app.med-os.in/onboarding and enable the Gen AI Addon in Settings → AI Addon.

This is just the beginning. The 10 features in version 1 cover the highest-value workflows. We're already planning voice-to-text consultation notes, per-doctor AI usage analytics, and bring-your-own-API-key for clinics that want to use their own Gemini quota. Stay tuned.