FindR Connect

Every second counts. Every face matters.

Comment

🔍 Inspiration

Every year, thousands of families are torn apart by the disappearance of a loved one. In India alone, over 100,000 children go missing annually — and a significant number are never found. The painful reality is that the first 48 hours are the most critical in any missing persons case, yet traditional reporting systems are slow, fragmented, and paper-heavy.

We asked ourselves: What if technology could close that gap?

FindR Connect was born from that question — a desire to put AI in the hands of communities and give every family a fighting chance. We were inspired by the power of facial recognition, the reach of social media, and the untapped potential of community-driven data to solve real-world humanitarian problems.

🛠️ How We Built It

FindR Connect is a cross-platform mobile application built on:

  • React Native & Expo — for a seamless iOS and Android experience
  • face-api.js & TensorFlow.js — powering on-device facial recognition using FaceNet models
  • Google Gemini API — for generative AI-assisted analysis and complex matching fallback
  • SQLite & AsyncStorage — enabling a local-first, offline-capable architecture

🤖 The Matching Algorithm

The core of FindR Connect is a multi-factor confidence scoring system. Match confidence $C$ is calculated as a weighted sum:

$$ C = 0.70 \cdot F + 0.10 \cdot A + 0.10 \cdot L + 0.05 \cdot G + 0.05 \cdot T $$

Where:

  • $F$ = Facial similarity score (FaceNet embedding cosine similarity)
  • $A$ = Age proximity score
  • $L$ = Last-seen location proximity score
  • $G$ = Gender match score
  • $T$ = Time proximity score (recency of report)

A match is flagged for review when $C \geq 0.75$, and auto-confirmed when $C \geq 0.90$.

📸 Reporting & Case Flow

  1. User submits a Missing or Found report with photo + details
  2. AI engine runs facial embedding extraction locally on-device
  3. New report is cross-referenced against the full database
  4. High-confidence matches trigger instant notifications
  5. Unresolved cases (48h+) are escalated to the automated social media awareness engine

📚 What We Learned

Building FindR Connect pushed us across multiple technical and human boundaries:

  • On-device ML is hard — Running FaceNet models efficiently inside a React Native environment required careful memory management and model quantization
  • UX for crisis situations is different — When a user is reporting a missing child, every extra tap is a failure. We learned to design for high stress, low patience scenarios
  • Data sensitivity demands respect — Working with photos and personal information of vulnerable individuals taught us to treat privacy not as a feature, but as a foundation
  • Community trust is the real algorithm — The best AI in the world is useless if people don't trust the platform enough to use it

⚔️ Challenges We Faced

1. Facial Recognition Accuracy

Getting reliable matches across varied lighting, angles, and image quality was our biggest technical hurdle. We addressed this by combining FaceNet embeddings with Gemini's generative analysis as a fallback, significantly improving robustness.

2. Offline-First Architecture

Missing persons cases don't wait for good WiFi. Designing a system that works reliably with intermittent or no connectivity — while keeping data synced when back online — required careful SQLite schema design and conflict resolution logic.

3. Simulating Real Infrastructure

Building a realistic CCTV network simulation and OTP authentication flow without access to actual government infrastructure meant we had to design convincing, extensible mock systems that could be swapped for real integrations.

4. Emotional Weight of the Problem

Perhaps the most unexpected challenge was the human one. Working on a platform meant to help families find missing loved ones kept us constantly grounded in why this work matters — and raised the bar for everything we built.

🌱 What's Next

FindR Connect is a foundation, not a finish line. Future directions include:

  • Integration with national missing persons databases (e.g., TrackChild, ICMIS)
  • Real CCTV API partnerships with municipal governments
  • Multilingual support for India's diverse linguistic landscape
  • A web portal for law enforcement and NGO partners

Built With

Share this project:

Updates