SETHU — AI Bridge for India's Missing Persons

SETHU means "bridge" in Sanskrit. AI-powered missing persons platform bridging families to their loved ones — with age progression across 30 years, real-time risk detection, and consent-first reunion.

Comment

Inspiration

In India, a child goes missing every 8 minutes.

Existing systems like TrackChild and TALAASH are police-only, have no AI, no age matching, and no public access. A person found in Chennai has no way to be matched to a report filed in Jharkhand. Families search for decades with nothing.

We kept asking — why hasn't anyone built a proper AI solution for this in India?

That question inspired SETHU.

We named it SETHU because Sethu (सेतु) means bridge in Sanskrit. This app is the bridge between a family and the person they have been searching for — sometimes for decades.

What it does

SETHU is a consent-first AI platform with four live engines powered by Claude:

1. 🧠 Age Progression Analysis Lost at age 9, searching at 35? AI generates how the person looks at every decade — bridging 30-year gaps impossible for conventional photo matching.

2. 🛡️ Risk Assessment Engine AI analyses every report for predator patterns — organ traffickers, loan sharks, abusive husbands, bonded labour contractors — and flags HIGH or CRITICAL risk before any match is shared.

3. 🔍 Sighting Match Analysis Citizen describes what they saw — AI cross-matches against case files accounting for age gap, location, and dialect. Gives confidence score with recommended police action.

4. 📋 Unidentified Person Profiler Hospital has an unidentified person who cannot speak. AI extracts origin clues from dialect fragments, physical markers, and behaviour — building a searchable identity from nothing.

The Consent Wall — Master Protection: Before any reunion, an NGO worker meets the found person alone and asks privately: "Are you safe? Do you want to go back? You can say no." If NO — case closes instantly. The reporter gets zero information. Even a perfect fake report by a trafficker gets them nothing.

4-Tier Access — No Aadhaar Required: Works for urban families, rural families with no smartphone, any citizen with a basic phone, and even zero-technology users via Childline 1098.

How we built it

  • Frontend: HTML, CSS, JavaScript — Canvas API for age-appropriate face visualisation
  • AI: Claude Sonnet via Anthropic API — four distinct prompt-engineered endpoints
  • Architecture: Single-page app with real live API calls — not mocked responses
  • Design: Built for police officers, NGO workers, and families — clear, fast, and accessible

Every AI feature calls Claude in real time. The risk assessment, age progression, sighting match, and case profiler are all live and working.

Challenges we ran into

1. Making AI the core, not a gimmick Designing features where AI was genuinely necessary — not just a chatbot wrapper. Each engine solves something that rule-based systems fundamentally cannot.

2. The predator problem A missing persons database could be weaponised by organ traffickers, loan sharks, and abusive partners. Every feature had to be redesigned with safety as the priority.

3. Excluding the most vulnerable Aadhaar-only excluded trafficking victims. Face AI excluded newborns. Smartphone-only excluded rural families. We built 4 access tiers to make the bridge truly universal.

4. The lying problem Any form can be lied on. A trafficker can tick "Lost/Wandered" and pass every check. Solution: stop catching liars at entry — make lying useless at exit. The Consent Wall means even a perfect lie gets the predator nothing.

Accomplishments that we're proud of

  • Built four live AI engines that call Claude in real time — not a single mocked response
  • Designed the Consent Wall — a safety feature where the found person's private YES or NO overrides everything, including a technically perfect fake report
  • Created 4-tier access so the platform works for people with no Aadhaar, no smartphone, and no documents
  • Designed a newborn protocol using footprint biometrics and DNA heel prick hash — because face AI completely fails for babies
  • Identified 7 specific predator attack vectors and blocked each one with a different safety layer
  • Named it SETHU — a name where every single feature is literally a bridge ## What we learned
  • The best AI applications solve problems that rules literally cannot solve
  • Safety must be designed from day one — not added later
  • The people who need a solution most are often the hardest to design for
  • Any form can be lied on — so catch predators at exit, not entry
  • A name matters — SETHU is not just a brand, it is the entire mission in one word ## What's next for SETHU — AI Bridge for India's Missing Persons
  • Integration with CCTNS and TrackChild government systems
  • DeepFace + GAN age progression model fine-tuned on South Asian faces
  • Partnership with Ministry of Home Affairs for national police portal rollout
  • NGO welfare check network across India
  • WhatsApp bot for rural zero-smartphone access
  • Newborn footprint digitisation at government hospitals
  • 28-state rollout with real-time cross-state matching
Share this project:

Updates