💡 Inspiration

Sri Lanka possesses an incredibly deep, vibrant cultural and historical backdrop spanning over 2,500 years. However, modern travelers and tech-savvy locals often lack an interactive, context-aware digital companion. We built "Lanka Heritage AI" for the World Product Day 2026 Hackathon to bridge this gap, acting as an intelligent spatial web agent that interprets environmental telemetry, synthesizes fluid audio guides on the fly, and runs advanced visual diagnostics.

⚙️ What it does

Lanka Heritage AI is a mobile-responsive web agent featuring:

  • Context-Aware Deep Linking: Instantly spins up historical contexts directly from URL path queries.
  • Scan Historical Path (GPS Telemetry): Tracks live coordinates and triggers reverse-geocoding for precise cultural itineraries.
  • Voice Input Pod & Audio Guide: Hands-free voice dictation combined with an asynchronous speech synthesis engine.
  • Vision Diagnostics: Allows users to snap or upload images of ancient ruins or crop leaf anomalies for real-time structural audits and agricultural triage.
  • Interactive Spatial Map: A custom Leaflet-driven map plotter mapping cultural landmarks and live agri-zones across Sri Lanka.

🛠️ How we built it

The core production stack is built natively on Replit using React.js (v19 modern hooks) and TypeScript for a type-safe environment. We used TailwindCSS for layouts and Lucide Icons for design. The intelligence layer orchestrates custom OpenAI API prompts, while Leaflet and OpenStreetMap handle the geo-visualization frame.

🛑 Challenges we overcame

  • Defeating Audio Deadlocks: Fixed mobile browser crashes by creating a cleanup cycle that kills active speech synthesis registers before rendering fresh state.
  • Preventing Camera Stream Leaks: Implemented hardware release hooks to forcefully stop camera tracks upon form cancellation, saving mobile device battery.

🏆 What's next for Lanka Heritage AI

We plan to introduce true offline edge-computed neural models for local languages (Sinhala/Tamil) and train computer vision models to transcribe ancient stone inscriptions directly from live media streams.

Built With

Share this project:

Updates