Sahej · सहज

Making sure no new mother in India misses the government help she's entitled to.

What is this? (in plain words)

When a baby is born in India, the government offers the mother real support — cash payments, free hospital care, free vaccines for the baby, and more. Hospital delivery and child vaccination now reach most families. But the cash benefits — the part that most directly helps a poor household — largely don't arrive in full or on time: a state-wide study found only about 1 in 10 mothers enrolled in the main maternity-cash scheme received all of what they were owed (see the charts below). Not because the money isn't there, but because:

• nobody tells them which schemes they actually qualify for,
• the forms and rules are confusing and scattered across departments, and
• each benefit has a deadline that quietly passes.

So thousands of crores meant for poor families go unused every year, and mothers go without help they had every right to.

Sahej fixes the navigation. It's a simple tool for the local government health worker — the ASHA — who already visits every new mother at home. The worker enters a few basic details about the mother and baby, and Sahej instantly shows:

• every benefit she's owed,
• how much money each one is,
• which documents she needs, and
• the date by which to claim each one —

laid out as a step-by-step checklist she can tick off over her routine visits, so nothing slips through the cracks.

In one line: Sahej turns a birth into a clear, dated checklist of every benefit the mother is owed — so the help actually reaches her.

A quick example

Enter a first-time mother in Bihar who delivered at a government hospital. Sahej replies: she's owed about ₹6,400 — claim the ₹1,400 delivery cash now, register the birth by day 21 (that unlocks another ₹2,000), get the baby's first vaccines, and so on. Change her state or situation and the answer updates automatically.

Why this approach works (the core insight)

For childbirth, a trusted person is already at the mother's door on a fixed schedule: ASHA workers make home visits on days 3, 7, 14, 21, 28 and 42 after birth (and are paid a small amount for each newborn they follow). Those visit days line up almost exactly with the benefit deadlines. So Sahej doesn't need to invent a new app habit or a new field force — it simply hands a smart checklist to someone who's already there, turning each visit into "here's exactly what this mother is owed today."

The problem, in three charts

Figures are from the sources listed below; the charts are generated from that data by tools/make_charts.py.

Hospital access is largely solved — the cash is where families fall off. Most mothers now deliver in a facility and most children are immunised; far fewer receive the full maternity cash they're entitled to.

The maternity-cash drop-off. Of mothers enrolled in PMMVY, under half received the first installment and only about 1 in 10 received all three.

And when the cash does come, it arrives months late — long after a newborn's costs have hit.

The money is allocated — roughly ₹1 lakh crore of welfare funds goes unspent every year. The gap is awareness, timing and follow-through. That's the gap Sahej closes.

What's in this build

File	What it is
PRODUCT_PLAN.md	The owner's plan: personas, the full scenario matrix, feature set, roadmap.
data/childbirth_schemes.json	The asset. Structured, sourced rules + all 36 states/UTs (LPS·HPS, opt-outs).
data/death_schemes.json	Second life event on the same engine: death registration, NFBS, widow pension, PMJJBY/PMSBY insurance claims, EPF/EPS/EDLI, BOCW death benefit, heir certificate.
engine.py	Pure-stdlib resolver: eligibility, blocking, claimed-tracking, urgency, documents, sensitive-mode, migrants, input validation. CLI + meta().
store.py	Accounts + storage (SQLite, stdlib): phone + PIN sign-in (PBKDF2-hashed, rate-limited), 30-day cookie sessions, per-worker caseload sync (last-write-wins, tombstone deletes), unguessable per-case share tokens.
serve.py	Zero-dependency hardened server: pages, API, auth routes, static/PWA assets, security headers (CSP, nosniff, frame-deny), traversal-safe.
web/landing.html	The story: problem, insight, how it works, coverage, partner CTA.
web/mother.html	The mother's own page (/m/<token>): read-only, Hindi-first, big type — her money, next step, documents and checklist. Shared by the ASHA over WhatsApp; no app, no login.
web/index.html	The ASHA tool (PWA): Today work plan across the caseload, application lifecycle (applied → received, stuck-payment detection + complaint generator), where-to-apply & grievance channels, voice intake (Hindi/English speech → filled form), life-event selector (childbirth / death in family), one-tap demo caseload, caseload backup/restore + CSV block report, EN⇄HI, docs checklist, alerts, share, offline support.
web/sw.js + manifest.webmanifest + icons	Installable app; shell cached offline, last plans available without signal.
api/index.py + vercel.json + requirements.txt	Vercel deploy: re-exports the handler as a serverless function, routes all paths to it, pulls the Postgres driver.
test_engine.py + test_catalog.py + test_server.py + test_store.py	234 checks: full scenario matrix, catalog/facets, HTTP routing/validation/security/leads/OTP, accounts/sessions/sync/content, consumer OTP login. CI runs them on every push.
tools/	Reproducible generators for the README charts and PWA icons.

Run it

python3 test_engine.py     # 85 scenario checks
python3 test_catalog.py    # 39 marketplace/facet checks
python3 test_server.py     # 61 HTTP/security/lead/OTP checks
python3 test_store.py      # 49 account/session/sync/content/OTP checks
python3 engine.py --state BR --birth-date 2026-06-01 --child-number 1 --child-sex girl \
    --area rural --mother-age 24                     # CLI report
python3 engine.py --birth-outcome stillbirth --state BR    # sensitive case
python3 serve.py           # http://localhost:8000 — landing at /, app at /app

No dependencies — Python 3.9+ standard library only (SQLite included). The web app installs to the home screen (PWA) and keeps working offline: the shell and each mother's last computed plan are cached on the phone.

Two interfaces, one engine

• The ASHA's tool (/app) — works with no account at all: the caseload lives in the phone's own storage, fully offline. Signing in (mobile number + PIN, in the caseload panel) adds a server copy, so the caseload survives a lost or changed phone — and unlocks the second interface:
• The mother's page (/m/<secret-token>) — every synced case gets a private link the ASHA can send on WhatsApp. The mother sees her own money, her next step and her documents in large Hindi type. Read-only, nothing to install, no login — the unguessable link is the key. Data stays minimal: the server stores only what the ASHA entered; PINs are salted-and-hashed, sessions are HttpOnly cookies, and wrong PINs rate-limit.

Deploy

The same serve.Handler runs three ways — pick one; no code changes.

Vercel + Postgres (Supabase or Neon — the hosted setup)

Serverless, scales to zero, free tier fits a pilot. The repo ships ready: api/index.py re-exports the handler, vercel.json routes every path to it and bundles web/ + data/, requirements.txt pulls the Postgres driver.

1. Database. Create a project on Supabase or Neon and copy the pooled connection string (Supabase: Settings → Database → Connection string → URI, pooling mode, port 6543; Neon: the host with -pooler).
2. Seed the catalog into the database (once, and after any data edit):

   DATABASE_URL='postgresql://…' python3 tools/migrate_to_db.py
   DATABASE_URL='postgresql://…' python3 tools/migrate_to_db.py --check

3. Import the repo into vercel.com (or vercel from this folder). Framework preset: Other — Vercel auto-detects the Python function.
4. Set env vars in the Vercel project → Settings → Environment Variables:
   • DATABASE_URL = the pooled string
   • SAHEJ_SECURE = 1 (marks the session cookie Secure)
5. Deploy. Every git push to main ships automatically.

What lives in the database: the scheme catalog + life-event knowledge bases (schemes, reference_docs), worker accounts and mother caseloads (workers, sessions, cases), and consumer mobile-number leads (leads). The schema is created on first request; tools/migrate_to_db.py seeds the content from data/*.json. Without DATABASE_URL the app still boots on local SQLite and serves the bundled JSON catalog, so CI and offline dev need zero configuration — the code path is identical, only the source differs.

Container (Railway, Fly.io, Render, a ₹300/mo VPS)

docker build -t sahej . && docker run -p 8000:8000 sahej

Bare metal

HOST=0.0.0.0 python3 serve.py — nothing to install (SQLite is used automatically when DATABASE_URL is unset). GET /healthz is the liveness probe.

Scenarios the engine handles

• All 36 states/UTs, with LPS/HPS JSY amounts and central-scheme opt-outs (e.g. West Bengal ≠ PMMVY).
• Parity & sex: 1st child, 2nd-girl (₹6,000), 2nd-boy (no PMMVY); girl-child state schemes.
• Delivery: public / private-empanelled / private / home; C-section; JSSK entitlements.
• Category & income: JSY gated to BPL/SC/ST in High-Performing states.
• Risk: premature / low-birth-weight → SNCU + extra visits; disability → UDID pointer.
• Sensitive outcomes: stillbirth, neonatal death, maternal death → death registration, NFBS, compassionate mode (no cheerful framing, only what applies).
• Migrants (delivered outside home state), missing Aadhaar/bank (hard blocker), govt employees, age & 270-day window.
• Journey state: already-claimed items, blocked-by-prerequisite, overdue / due-soon urgency.

Schemes encoded

PMMVY, JSY, JSSK (incl. SNCU), Birth Registration, Death/Stillbirth Registration, Universal Immunization, RBSK, NFBS (survivor benefit), a disability/UDID pointer, and representative state schemes (Tamil Nadu, Odisha, Madhya Pradesh, West Bengal).

Honesty by design

This domain punishes hallucinated rules — a wrong "you qualify" costs a mother a day's wage. Every rule carries confidence, source_urls, and a needs_verification flag, surfaced in the UI. Amounts/conditions are research-grade drafts — confirm against current Government Orders before real use. Not medical or legal advice.

Evidence & sources

Maternity cash reaches few mothers in full — and late

• PMMVY state-wide evaluation: only ~10% received all three installments; average 90 / 113 / 137-day delays — BMC Pregnancy & Childbirth (2025) — PMC · Springer
• PMMVY scale (40.8M mothers paid, ₹19,160 cr since 2017) — PIB, 2025
• PMMVY scheme details & 270-day window — UMANG / MoWCD · Vikaspedia

Coverage of care & immunisation (the denominators in the first chart)

• Institutional delivery ~89%; full child immunisation ~77% — NFHS-5 — PIB · PIB Phase-II

The money is there — it goes unspent

• ~₹1 lakh crore of welfare funds unspent — Policy Circle
• Underutilisation & awareness gaps (incl. PMMVY) — Drishti IAS

The delivery rail — why ASHA visits are the trigger

• Home-Based Newborn Care visit schedule (days 3/7/14/21/28/42; ₹250 per newborn) — NHM HBNC

Existing players — why a proactive, last-mile tool is the gap

• myScheme — the government's scheme discovery portal — myscheme.gov.in
• Haqdarshak — assisted-tech with human field agents — Acumen case study

The PMMVY completion and delay figures come from a state-wide study, not a national census — they indicate the size of the gap, not a precise national average. All figures are research-grade; verify against current Government Orders before operational use.

Beyond this MVP

WhatsApp-native delivery, supervisor/block dashboards on top of the synced caseloads, auto-submission to government portals, Bhashini for deeper language coverage, and the next life events (disability, job loss) on the same engine.