MedCore

Hospital Information System for Indian Healthcare

A full-stack, monorepo HIS covering the patient journey from appointment to discharge — clinical, operational, financial, HR, and engagement workflows in one typed codebase.

Live Demo · Features · Architecture · Testing · Deployment · Roadmap · Commercial

---

Overview

MedCore is a hospital information system built as a TypeScript monorepo. It spans outpatient and inpatient care, emergency, surgery, pharmacy, lab, blood bank, billing with GST and Razorpay, HR and payroll, patient engagement, and a React Native patient app. The web dashboard, API, and mobile app share validation schemas and types end to end.

The project is under active development. A live demo instance runs at medcore.globusdemos.com and is exercised by the Playwright E2E suite on every push.

At a glance

Tests passing	~2,700+ active (api unit/integration + web component + shared/contract + smoke) across unit / contract / component / integration / RBAC layers; +1,873 DB-integration cases skipped without DATABASE_URL_TEST
E2E	40 Playwright specs (e2e/*.spec.ts); release.yml runs full Chromium + WebKit on workflow_dispatch
Accessibility	axe-core scan across 19 routes (public marketing + login + 12 dashboard pages), WCAG 2.1 AA, per-page contrast budgets
API routers	80+ (incl. AI: triage, scribe, radiology, KPIs, agent-console, sentiment, fraud, doc-QA, letters; plus ABDM, FHIR, HL7 v2 inbound, insurance claims, chart-search, patient-data-export)
Prisma models	155+
Prisma migrations (production)	18, all applied via migrate deploy
Triage languages	8 (English, Hindi, Tamil, Telugu, Bengali, Marathi, Kannada, Malayalam)
CI workflows	6 (test.yml per-push gate, release.yml full e2e, codeql.yml, ai-eval-nightly.yml, load-test-nightly.yml, update-visual-baselines.yml)
Demo URL	https://medcore.globusdemos.com

---

Feature Catalog

AI Features

All AI features run on Sarvam AI (sarvam-105b), an Indian LLM provider, ensuring data residency within India for DPDP Act compliance. Speech-to-text uses Sarvam ASR (saaras:v3). The AI layer is observable (every LLM call logged with latency, tokens, and truncated prompt), retry-resilient (withRetry with exponential back-off), and tested via a Vitest eval harness with gold-standard fixtures.

• AI Triage Chatbot (/dashboard/ai-booking) — multi-turn symptom collection in 8 Indian languages (English, Hindi, Tamil, Telugu, Bengali, Marathi, Kannada, Malayalam) with localised UI strings + 10-symptom chip row per language. Deterministic red-flag detection (cardiac, stroke, respiratory, bleeding, suicidal ideation, obstetric, neonatal, Hindi/Devanagari phrases) fires before the LLM, triggering an immediate emergency screen with call-112 instructions. After 4+ exchanges the LLM assesses the complaint, with specialty matching reconciled against a curated SNOMED-CT subset (119 concepts incl. Hindi synonyms) for explainability. Supports booking on behalf of a dependent (child, elderly parent). Patients can skip AI triage and go straight to manual booking.
• Agent Console (/dashboard/agent-console) — when a triage chat needs a human, the call-centre agent sees the full transcript, AI-extracted SOAP, top-3 doctor matches and red-flag annotations on a single screen. One-click hand-off, audit-logged, real-time updates over Socket.IO.
• AI KPI Dashboards (/dashboard/ai-kpis) — PRD-grade KPIs for both AI flagship features: triage routing accuracy, AI-flow CSAT, scribe time-saved, doctor NPS, drug-interaction catches, scribe-vs-manual med-error rate. CSV export and per-tenant filtering. Backed by apps/api/src/services/ai/kpi-metrics.ts with explicit unavailable: true semantics for KPIs that need data we don't have yet (so nobody mistakes an empty value for a computed zero).
• AI Radiology Report Drafting (/dashboard/ai-radiology) — upload an imaging study, AI drafts an impression + per-finding confidence + suggested follow-ups, radiologist approves or amends with a strict HITL workflow. DICOM-ready (dicom-parser), with click-to-highlight region overlays. The route never produces a FINAL report on its own — approveReport() is the only way to move to FINAL.
• AI Scribe (/dashboard/scribe) — ambient speech-to-text during consultation via Sarvam ASR or Web Speech API. Transcripts stream to the API every 5 final utterances; once 3+ entries accumulate, the LLM generates a structured SOAP draft (Subjective / Objective / Assessment / Plan) with ICD-10 code suggestions, CPT codes, and per-section confidence scores. Doctors can review by section (Accept / Edit / Reject) using voice commands ("accept subjective", "edit plan", etc.). On sign-off the note is written to the EHR consultation record, and draft lab orders and referrals are auto-created from the SOAP plan.
• Drug Safety Check — runs automatically inside the Scribe transcript endpoint on every SOAP draft. Two-layer architecture: fast deterministic layer checks ~15 curated high-risk pairs (warfarin + NSAIDs, SSRI + MAOI, sildenafil + nitrates, etc.), allergy cross-reactivity families, condition contraindications, paediatric contraindications, and renal/hepatic dosing flags. The LLM catches anything outside the curated list. Alerts are severity-coded CONTRAINDICATED → SEVERE → MODERATE → MILD with generic alternatives suggested. A DrugAlertBanner in the Scribe UI blocks sign-off on CONTRAINDICATED alerts until the doctor explicitly acknowledges.
• Medication Adherence Bot (/dashboard/adherence) — enroll a patient's prescription into a reminder schedule. A 15-minute background scheduler checks for due medications and sends AI-personalised WhatsApp/SMS reminders in the patient's preferred language.
• AI Lab Report Explainer (/dashboard/lab-explainer) — generates a plain-language explanation of lab results for the patient, flagging abnormal values. Explanations are held in a HITL approval queue; a doctor reviews and approves before the explanation is sent to the patient.
• AI Letter Generator (/dashboard/letters) — one-click generation of referral letters and discharge summaries from structured clinical data. Letters are editable and printable directly from the browser.
• No-Show Prediction (/dashboard/predictions) — rule-based model (7 features: historical no-show rate, lead time, day of week, hour of appointment, new-patient flag, recent no-show flag, appointment type) predicts per-appointment risk. Batch predictions run nightly; the dashboard shows high-risk appointments for proactive follow-up.
• ER Triage Assist (/dashboard/er-triage) — MEWS scoring (respiratory rate, O₂ sat, heart rate, systolic BP, temperature, consciousness) plus AI-suggested ESI triage level with clinical rationale.
• Pharmacy Inventory Forecasting (/dashboard/pharmacy-forecast) — forecasts stock requirements for the next 30/60/90 days based on dispensing history, with AI-generated procurement insights.
• AI Analytics (/dashboard/ai-analytics) — tabbed dashboard showing triage session volume and conversion rates, scribe session counts, sign-off rates, average edit counts, and doctor-edit heatmaps.
• Knowledge Base (RAG) — PostgreSQL full-text search (to_tsvector / plainto_tsquery) over a KnowledgeChunk table seeded from ICD-10 codes, medicine catalogue, and clinical protocols. Retrieved context is injected into every LLM prompt to ground responses in hospital-specific data without requiring pgvector.

Shipped this sprint (April 2026)

• PRD §3 / §4 / §7 closure (Apr 24) — Phase-2 regional languages (Tamil, Telugu, Bengali, Marathi, Kannada, Malayalam) joined English + Hindi for triage; SNOMED-CT 119-concept curated subset with Hindi synonyms; agent console for call-centre hand-off; AI-KPI dashboards with CSV export; AI Radiology drafting (DICOM-aware + region overlays); 317-word medical-vocabulary boost list (later removed when AssemblyAI/Deepgram were dropped, see below); patient data export under DPDP §11 right-to-portability; multi-tenant onboarding with self-serve provisioning.
• AssemblyAI + Deepgram removed (Apr 25) — both processed audio in US-region data centres, violating PRD §3.8 / §4.8 ("India-region for all PII and PHI"). ASR is now Sarvam-only with a noisy factory error if ASR_PROVIDER is set to anything else. Re-introduce only under a DEPLOYMENT_REGION runtime gate if MedCore ever expands outside India.
• Prompt registry with versioning + rollback (/api/v1/ai/admin/prompts) — every LLM prompt (TRIAGE_SYSTEM, SCRIBE_SYSTEM, DRUG_SAFETY, etc.) lives in the Prompt table as a row per version. Admins can POST a new (inactive) version, activate it with a one-click API call, and one-shot roll back. 60-second in-memory cache; falls back to the hardcoded PROMPTS constant if the DB is unreachable. Every mutation audit-logged.
• Claims auto-draft from SOAP (POST /api/v1/ai/claims/draft/:consultationId) — takes the AI Scribe output (SOAP + ICD-10 + CPT codes) and the consultation invoice, emits a pre-filled InsuranceClaim2 draft. Reception reviews in 30 seconds. Includes a denial-risk predictor and a machine-replayable auto-fix endpoint.
• Speaker tagging in Scribe transcripts — live scribe UI assumes alternating speakers; doctors can manually relabel any utterance as DOCTOR / PATIENT / ATTENDANT (acoustic diarization deferred until an India-region provider is available).
• Massive QA-issue closure (Apr 26–27) — 70+ GitHub issues closed across two sweeps: RBAC hardening (RECEPTION + DOCTOR boundaries), payroll math (FY-2026 ESI ceiling + Net-Pay parity), pharmacy validation, dark-mode contrast across 22 pages, friendly outage page, scheduled tasks for stale surgery + overdue complaint auto-actions, and the new revenue / elapsedMinutes / formatDate helpers that retired half-a-dozen long-standing math + timezone bugs. See docs/archive/SESSION_SNAPSHOT_2026-04-27.md for the full per-issue breakdown.

Deeper architecture, observability, retry, and HITL details live in docs/AI_ARCHITECTURE.md.

Compliance & Interoperability

• ABDM / ABHA Gateway (/api/v1/abdm) — ABHA address and 14-digit ABHA number verification, link / delink to MedCore patient records, consent-artefact creation (ABDM §5 CM flow), and CareContext discovery for Health Information Providers. Validates ABHA address format (handle@domain) and 14-digit ABHA numbers against the NN-NNNN-NNNN-NNNN pattern. Every gateway call is audit-logged.
• FHIR R4 export (/api/v1/fhir) — read-only export of Patient, Encounter, and Patient/:id/$everything bundles. Responses use application/fhir+json content type per FHIR R4 §3.1.6. Errors surface as OperationOutcome resources. Role-gated so that only clinicians and the patient themselves can read. Every resource read is audit-logged (FHIR_PATIENT_READ, FHIR_PATIENT_EVERYTHING, etc.).
• Insurance / TPA Claims (/api/v1/claims) — pre-authorisation lifecycle, claim submission, document attachments, and status-event timeline. Works alongside the existing /api/v1/preauth workflow; status events are persisted for audit and SLA tracking.
• Jitsi deep integration — the existing /api/v1/telemedicine router now supports a waiting-room lifecycle (PATIENT_WAITING → ADMITTED / DENIED), with dedicated join/admit endpoints and realtime state propagation.
• Consent & retention — ConsentArtefact model captures patient consent scope, purpose, and expiry. An audio-retention scheduler enforces consultation-audio retention windows.

Intelligence Layer

A cross-cutting AI substrate that every feature above plugs into. The pieces:

• Sarvam AI (sarvam-105b, India region, DPDP-compliant) — single LLM vendor for every feature. ASR uses saaras:v3. All calls flow through apps/api/src/services/ai/sarvam.ts.
• RAG + Postgres FTS — KnowledgeChunk table with to_tsvector / plainto_tsquery. No pgvector required. Seeded from ICD-10, medicine catalogue, and clinical protocols; enriched by the ingest pipeline (below).
• Fire-and-forget ingest pipeline (services/ai/rag-ingest.ts) — SOAP notes, lab results, prescriptions, and uploaded patient documents are chunked (~800 chars, paragraph-aware) and upserted into KnowledgeChunk from the originating route handler. tesseract.js handles OCR for image uploads; pdf-parse handles text extraction from PDFs. Ingest failures never break the request path.
• ML no-show predictor (services/ai/no-show-predictor.ts + services/ai/ml/logistic-regression.ts) — 7-feature logistic-regression model over historical appointment data. Nightly batch prediction populates /dashboard/predictions.
• Holt-Winters pharmacy forecast (services/ai/ml/holt-winters.ts) — triple-exponential-smoothing time-series forecast of medicine demand with seasonality. Drives 30/60/90-day procurement insights on /dashboard/pharmacy-forecast.
• Ambient chart search (/api/v1/ai/chart-search) — free-text search over a patient's own ingested chart (notes, labs, prescriptions, uploaded docs) with LLM-synthesised answer. Currently ranked purely by Postgres FTS — see Known Follow-ups for the rerank roadmap.
• Observability, retry, HITL — every LLM call goes through logAICall (JSON logs with feature, tokens, latency) and withRetry (3 attempts, exponential back-off, degrades to AIServiceUnavailableError + HTTP 503 on exhaustion). Patient-facing AI output (lab explanations, adherence reminders) goes through a doctor-approval queue before reaching the patient.
• Eval harness — Vitest + gold-standard fixtures under apps/api/src/test/ai-eval/. Runs locally and gates regressions on triage red-flag recall, SOAP note accuracy, and drug-safety alerts.

Clinical

• OPD — appointments, walk-in queue with token generation (DB unique constraint, race-tested), live queue updates over Socket.IO, vulnerability flagging for at-risk patients.
• Prescriptions — drug interaction checks, renal dose calculator, scannable QR codes (real PNG generated with qrcode, decodable via jsqr), public /verify/rx/[id] verification page, PDF export via pdfkit.
• Lab — multi-test orders, result entry, delta flag against previous values, CRITICAL panic-value alerts, QC with Levey-Jennings charts, TAT tracking.
• Admissions / IPD — ward and bed management, nurse MAR, isolation, bed occupancy, discharge summary PDF.
• Emergency — 5-level triage, MEWS/GCS/RTS scoring, live ER board via ER:update Socket.IO events, MLC tracking.
• Surgery / OT — pre-op checklist enforced server-side, intra-op timing, PACU, SSI tracking, OT calendar.
• Maternity & Pediatrics — antenatal workflow with ACOG risk scoring and SVG partograph, pediatric growth charts, India UIP immunization schedule.
• Blood Bank — donor registry, ABO/Rh matching, component separation, unit reservations.
• Ambulance — dispatch with status transitions, fleet and fuel logs.
• Telemedicine — Jitsi video sessions with in-call chat and prescription creation.

Operations

• Inventory — reorder thresholds with scheduled auto-PO generation.
• Purchase orders — approve/receive workflow, partial GRN, three-way match.
• Suppliers — contracts, GST details, performance tracking.
• Assets — depreciation, calibration, warranty alerts, QR tags, dispose workflow.
• Visitor management — check-in with a 2-per-patient cap, printable passes, blacklist.
• Health packages — preventive bundles with validity and family sharing.

Finance

• GST-aware invoicing with CGST + SGST split, amount-in-words, PDF via pdfkit.
• Razorpay integration — server-side verifyPayment (fail-closed in production) plus a real webhook handler at POST /api/v1/billing/razorpay-webhook with HMAC-SHA256 raw-body verification, idempotency via Payment.transactionId @unique, and amount cross-check against the Razorpay REST API.
• Insurance / TPA — pre-authorization and claim workflow.
• Refunds, credit notes, discounts (flat / percentage) with threshold-based approvals.
• Expenses and budgets with category tracking and monthly variance.

HR and People

• Shift roster grouped by morning / afternoon / night.
• Leaves — balance tracking, approval workflow, 6 leave types, calendar view.
• Payroll — basic + allowances + overtime with approval and pay-slip generation.
• Seven roles — ADMIN, DOCTOR, NURSE, RECEPTION, PATIENT, PHARMACIST, LAB_TECH. A permissions matrix test exercises 178 role/endpoint assertions.
• Certifications and holiday calendar (Indian public holidays template).

Patient Engagement

• Feedback with 5-star ratings, NPS, sentiment analysis.
• Complaints with SLA due-at calculation and escalation.
• Internal chat over Socket.IO with reactions, pinning, mentions, department channels.
• Notifications — 13 notification types × 4 channels (WhatsApp / SMS / Email / Push) with templated messages, quiet-hours defer, drainScheduled cron that picks up both scheduled and NULL scheduledFor rows, and retry-once on failure.

Mobile (React Native + Expo SDK 53)

• Patient app — appointments, live queue over Socket.IO, prescription viewer, billing tab with native Razorpay checkout or WebView fallback, push notifications via expo-notifications.
• Patient AI screens — AI triage chat, medication adherence tracker, and plain-language lab explanations on the mobile tab bar (apps/mobile/app/ai/), calling the same API endpoints as the web dashboard.
• Doctor-lite app — workspace, patients, prescriptions.
• 401 refresh interceptor, env-driven API URL via expo-constants, EAS build profiles for dev/preview/production.

Security

• Persistent auth state — 2FA temp tokens and password-reset codes are stored in the database, not in-memory maps, so they survive restarts.
• TOTP 2FA implemented with pure Node crypto (no external library).
• Refresh-token rotation with replay detection; JWTs include jti to avoid same-second collisions.
• File uploads — row-level ACL, HMAC-signed URLs, magic-byte MIME sniffing that rejects an executable renamed to .jpg, 10 MB cap.
• Audit log on every mutation with CSV export.
• Rate limits — 600 req/min global, 30 req/min on auth endpoints.

Accessibility and i18n

• WCAG 2.1 AA baseline; axe-core CI gate with per-page budget overrides. Hard-fails on button-name, select-name, label, and image-alt.
• English and Hindi translations across 374 keys covering 10 dashboard pages. <html lang> switches reactively.

PDF Generation

Server-side PDFs via pdfkit for prescriptions (with embedded PNG QR), invoices (GST breakdown, amount-in-words), and discharge summaries. Routes branch on ?format=pdf so the HTML print flow still works for backward compatibility.

---

Tech Stack

Layer	Tools
Runtime	Node.js 20, TypeScript
API	Express 4, Zod validation, Socket.IO server
Database	PostgreSQL 16, Prisma 6 (migrations, 151 models)
Web	Next.js 15 (App Router), React 19, Tailwind CSS v4, Zustand, socket.io-client
Mobile	React Native, Expo SDK 53, expo-router, expo-notifications, expo-constants; patient AI screens (triage, lab explainer, adherence)
Auth	JWT with refresh rotation, bcrypt, TOTP 2FA (pure Node crypto)
Payments	Razorpay SDK, raw-body HMAC-SHA256 webhook verification
AI / LLM	Sarvam AI sarvam-105b (OpenAI-compatible, India region), Sarvam ASR saaras:v3
Document ingest	tesseract.js (OCR for image uploads), pdf-parse (text extraction from PDFs)
Interop	FHIR R4 export, ABDM / ABHA gateway client, Insurance / TPA claim workflow
PDF / QR	pdfkit, qrcode, jsqr (verification)
Testing	Jest, Supertest, Playwright, axe-core, Vitest (LLM eval harness)
Monorepo	Turborepo, npm workspaces
Ops	PM2, systemd, nginx, Let's Encrypt, Docker (Postgres), pg_dump backups
CI	GitHub Actions — test.yml per-push gate (typecheck, lint, API + web tests, npm-audit, migration-safety, web-bundle, auto-deploy); release.yml full Playwright (Chromium + WebKit) on workflow_dispatch; codeql.yml, ai-eval-nightly.yml, load-test-nightly.yml, update-visual-baselines.yml

---

Architecture

medcore/
├── apps/
│   ├── api/          Express API — routers, services, Socket.IO gateway
│   ├── web/          Next.js 15 dashboard (App Router, React 19)
│   └── mobile/       React Native (Expo) patient and doctor-lite apps
├── packages/
│   ├── shared/       Zod schemas, shared types (end-to-end type safety)
│   └── db/            Prisma schema, client, seeds, migrations
├── e2e/              Playwright specs (40 spec files; explicit-invocation only — see docs/TEST_PLAN.md §3 Layer 5)
├── scripts/          deploy.sh, backup, health-check, migration helpers
└── docs/             PRD, ARCHITECTURE, DEPLOYMENT, MIGRATIONS, AI_ARCHITECTURE, TEST_PLAN, 68 Playwright screenshots

Key decisions

• End-to-end type safety. The same Zod schemas validate API requests and generate TypeScript types consumed by the web and mobile apps.
• Prisma-first data modeling. Production uses prisma migrate deploy — never db push. The current migration history covers initial schema, auth persistence tables, PHARMACIST/LAB_TECH roles, Razorpay + push-token drift, marketing enquiry, AI feature tables, triage consent fields, AI-feature model expansion, and the ABDM / insurance / Jitsi / RAG compliance layer. Full policy, hand-crafting rules, and the .prisma-models*.md proposal pattern live in docs/MIGRATIONS.md.
• Socket.IO for realtime. Live OPD queue, ER board, chat, admissions.
• Fail-closed payments. Razorpay signature mismatch returns 400 in production; webhooks use raw-body verification and idempotency on Payment.transactionId.
• Row-level file ACLs with HMAC-signed URLs and magic-byte sniffing.
• Sarvam AI for DPDP compliance. All LLM calls route to Sarvam AI's India-region endpoint (api.sarvam.ai), satisfying the Digital Personal Data Protection Act's data-residency requirement. The SARVAM_API_KEY env var replaces any previous cloud LLM key.
• RAG without pgvector. The knowledge layer uses PostgreSQL full-text search (to_tsvector/plainto_tsquery) over a KnowledgeChunk table, avoiding the need for a Postgres extension and keeping the deployment footprint unchanged.
• Human-in-the-loop for clinical output. Lab explanations require doctor approval before reaching the patient. Scribe SOAP notes require explicit doctor sign-off. Drug CONTRAINDICATED alerts block sign-off until acknowledged.

---

Testing

MedCore layers its tests so each tier tests a different boundary:

Layer	Count	What it covers
Unit	~550	Helpers, validators, utilities, notification channel adapters, red-flag detection (51 cases), ML primitives (Holt-Winters, logistic regression), ABDM client, FHIR resource builders, consent service
Contract	~140	Zod request/response schemas between API and web
Smoke	30	Fast sanity pass across critical routes
Web	~420	React component and page-level tests, including the new AI dashboard pages
Integration	~900	Full HTTP through Express + Prisma against a real Postgres. Includes concurrency, realtime delivery, permissions matrix, auth edges, 2FA, notification channel shapes, Razorpay webhook, AI triage / scribe / chart-search / letters / predictions / report-explainer / adherence / er-triage / pharmacy / knowledge / transcribe, insurance claims, and telemedicine-deep (waiting room)
Mobile	30	React Native render / logic tests across the patient AI screens
AI eval	Vitest harness	Gold-standard fixtures under apps/api/src/test/ai-eval/; gates regressions on triage red-flag recall and SOAP accuracy
Total	~2,700+

In addition:

• 40 Playwright E2E specs under e2e/. Per the explicit-invocation policy, Playwright runs only via scripts/run-e2e-locally.sh locally or the release.yml workflow_dispatch gate in CI — never on per-push.
• axe-core a11y scan across 19 routes (7 public marketing + login + 12 dashboard pages) with per-page color-contrast budgets.

Running tests locally

The unified runner mirrors every per-push CI gate from test.yml:

# Full guide: docs/LOCAL_TESTING.md
scripts/run-tests-locally.sh --quick            # 3-5 min, no DB
scripts/run-tests-locally.sh                    # 5-7 min, default tier
scripts/run-tests-locally.sh --with-integration # default + integration suite
scripts/run-tests-locally.sh --with-e2e         # default + Chromium e2e (~15-20 min)
scripts/run-tests-locally.sh --with-e2e=both    # mirrors release.yml (Chromium + WebKit)

For Playwright e2e against a local Postgres-API-Web topology see scripts/run-e2e-locally.sh (full guide: docs/LOCAL_E2E.md).

Individual layers are still available via the workspace scripts:

npm --prefix apps/api run test:unit
npm --prefix apps/api run test:contract
npm --prefix apps/api run test:smoke
npm --prefix apps/api run test:integration   # requires Postgres
npm --prefix apps/web run test
npx playwright test                          # local Playwright

GitHub Actions wires six workflows: per-push test.yml (typecheck + lint + npm-audit + migration-safety + web-bundle + api-tests + web-tests + auto-deploy), explicit-only release.yml (full Chromium + WebKit Playwright), codeql.yml, ai-eval-nightly.yml, load-test-nightly.yml, and update-visual-baselines.yml.

---

Screenshots

All 68 Playwright screenshots live in docs/screenshots/. A curated selection is below.

Role dashboards

Admin	Doctor	Nurse

OPD and patient

Appointments	Live queue	Prescriptions

Clinical

Emergency / triage	Surgery	Lab

Operations and finance

Billing	Purchase orders	Analytics

View all 68 screenshots

Auth: login · register · forgot password

Dashboards: admin console · calendar · doctor workspace · nurse workstation

OPD: walk-in · token display · patient list · immunization schedule · vitals · controlled substances

IPD: wards · admissions · medication dashboard · census

Acute: OT · telemedicine · antenatal · pediatric · referrals

Diagnostics: medicines · pharmacy · lab QC · blood bank · ambulance · assets

Finance: refunds · payment plans · pre-auth · discount approvals · packages · suppliers · expenses · budgets

HR: duty roster · my schedule · leave management · my leaves · leave calendar · holidays · payroll · certifications

Admin: users · doctors · schedule · reports · scheduled reports · audit log

Engagement: notifications · broadcasts · feedback · complaints · chat · visitors

---

Quick Start

# Clone
git clone https://github.com/Globussoft-Technologies/medcore.git
cd medcore

# Install dependencies
npm install

# Start Postgres via Docker
docker run -d --name medcore-postgres \
  -e POSTGRES_USER=medcore -e POSTGRES_PASSWORD=medcore_dev \
  -e POSTGRES_DB=medcore -p 5433:5432 postgres:16-alpine

# Configure environment
echo 'DATABASE_URL="postgresql://medcore:medcore_dev@localhost:5433/medcore?schema=public"' > .env
cp apps/api/.env.example apps/api/.env
# Add your Sarvam AI key (required for all AI features)
echo 'SARVAM_API_KEY=your_key_here' >> apps/api/.env

# Apply migrations (do not use db push)
npx prisma generate --schema packages/db/prisma/schema.prisma
npx prisma migrate deploy --schema packages/db/prisma/schema.prisma

# Seed realistic data
npx tsx packages/db/src/seed-realistic.ts

# Start dev servers
npm run dev
#   Web: http://localhost:3000
#   API: http://localhost:4000

Mobile app:

cd apps/mobile && npx expo start

---

Demo Accounts

Role	Email	Password
Super Admin	[email protected]	superadmin123
Admin	[email protected]	admin123
Doctor	[email protected]	doctor123
Nurse	[email protected]	nurse123
Reception	[email protected]	reception123
Pharmacist	[email protected]	pharmacist123
Lab Tech	[email protected]	labtech123
Patient	[email protected]	patient123

Try them on the live demo.

---

Deployment

Production runs on a single Ubuntu 22.04 host behind nginx with Let's Encrypt:

nginx (443)
  ├─ medcore.globusdemos.com    → Next.js (:3200)
  └─ /api                        → Express API (:4100)
                                    └─ Docker PostgreSQL (:5433)

PM2: medcore-api, medcore-web (systemd auto-restart)
Cron: daily pg_dump @ 02:00 + health check every 5 min

scripts/deploy.sh

The deploy script ships code, runs prisma migrate deploy, restarts PM2, and runs a post-deploy health check. Seeding production is gated behind an explicit opt-in:

# Dangerous — only for a fresh environment. Wipes the database.
ALLOW_PROD_SEED_RESET=YES_I_WILL_WIPE_THE_HOSPITAL ./scripts/deploy.sh --seed

Without the exact magic string the script refuses to seed.

Backups

Daily gzipped pg_dump with 30-day retention. Restore rehearsal has been verified — 8 sampled tables match the source dump byte-for-byte.

---

Accessibility

• Axe-core CI gate with per-page color-contrast budgets.
• A single parameterized axe-core spec (e2e/a11y.spec.ts) sweeps 19 routes (7 public marketing + login + 11 dashboard pages including admin-console).
• Hard-fails on button-name, select-name, label, and image-alt rules.
• 20+ aria-labels added to icon-only controls.
• Heading-order budget currently 13 nodes (raised from 10 in e6f6d24 while shared chrome a11y consolidation is in flight).

---

Internationalization

• English and Hindi across 374 translation keys.
• 10 dashboard pages wired to useTranslation.
• <html lang> switches reactively with the active locale.
• A dev-only I18N.md documents the "add both languages in the same PR" rule.

---

Known Follow-ups

This is an honest list. Nothing below is hidden in a marketing footnote.

• HIPAA / ABDM compliance has not been third-party audited. The codebase implements relevant controls (audit log, signed URLs, encryption at rest via Postgres, consent artefacts, FHIR-native error envelopes) but the certifications are not in place.
• ABHA/ABDM health ID linking is roadmap (GAP-T13). Shipped — verify / link / delink / consent / CareContext endpoints at /api/v1/abdm.
• HL7 / FHIR export is planned, not built. FHIR R4 shipped at /api/v1/fhir (Patient, Encounter, $everything). HL7 v2 legacy export (for older lab analyzers and LIS gateways) is still not built.
• Multi-tenant / multi-branch is roadmap. Scaffolding (middleware/tenant.ts, tenantAsyncStorage) is in place; the full migration plan lives in .prisma-models-tenant.md and has not been applied — no Tenant table yet.
• The mobile doctor-lite app is intentionally a subset of the web workspace.
• Jitsi tele-consult deep integration (screen share, waiting room) is deferred (GAP-S14). Waiting room shipped (PATIENT_WAITING → ADMITTED / DENIED). Screen share and in-call recording remain deferred.
• Insurance billing claims API integration is deferred (§7-8). Scaffold shipped at /api/v1/claims (pre-auth lifecycle, document attachments, status-event timeline). Direct payer-specific connectors (Star, HDFC Ergo, etc.) are still partner-gated.
• AI ambient chart search is live at /api/v1/ai/chart-search but currently ranks purely on Postgres FTS ts_rank. A cross-encoder / LLM rerank layer on top of the top-K FTS hits would materially improve precision for long patient histories — not yet built.

---

Contributing

See CONTRIBUTING.md for the full guide, including the Prisma migration policy — all schema changes must ship as a prisma migrate dev migration file. prisma db push is not permitted against any shared environment.

Quick path:

git checkout -b feat/your-feature
scripts/run-tests-locally.sh --quick     # 3-5 min: typecheck + lint + audit + migration-safety + bundle
scripts/run-tests-locally.sh             # 5-7 min: + api/web vitest tiers
git commit -m "feat: add awesome thing"
git push origin feat/your-feature

Standards:

• TypeScript strict mode, Zod at every API boundary.
• Audit logging on every mutation.
• Tailwind utilities; avoid new CSS files.
• No new runtime npm dependencies without discussion.
• Both English and Hindi keys added for any user-facing string.

---

Commercial Licensing

MedCore is proprietary software owned by Globussoft Technologies. The repository is published for transparency and evaluation. For:

• Deploying MedCore at a hospital
• White-labeling or OEM arrangements
• Custom integrations (ABDM, HL7/FHIR, specific lab analyzers, insurance APIs)
• Training and SLA-backed support
• Reseller / implementation partnerships

open an issue or contact Globussoft Technologies.

---

License

Proprietary. See LICENSE. Commercial licensing inquiries welcome.

---

Live Demo · GitHub · Contact

Built in India for Indian hospitals.