The goal is to build a GPT-powered service that fetches live railway-station data—platform numbers and upcoming train timings—exclusively from publicly available websites, then serves the answers instantly on both web and mobile front ends. How it should work • The back-end agent receives a user query such as “Next three departures from Mumbai CST” or “Which platform for Train 12002 at Agra Cantt at 18:00?”. • It scrapes or calls the chosen public sites in real time, cleans the results, and feeds accurate, structured information into GPT so responses are conversational yet precise. • A lightweight REST (or GraphQL) layer exposes the data to a React / Flutter / PWA interface—whichever you find fastest for simultaneous web and mobile delivery. Key requirements • Real-time data refresh with minimal lag; caching no longer than 60 seconds for time-sensitive endpoints. • Clear separation between the scraper/parser, the GPT prompt logic, and the UI so each piece can be swapped or upgraded independently. • Include polite fallback replies for missing or delayed data instead of showing errors. • Code, setup scripts, and a one-page deployment guide must run on a standard Linux VPS. Acceptance criteria 1. Type “/status Delhi” in the demo chat and receive at least five upcoming departures with correct platform numbers matching the source site. 2. The same request from a mobile browser returns identical data formatted responsively. 3. Source URLs and timestamps are logged for every answer and can be exported as CSV. I’ll provide a test server and sample source links once we start; you take it from first commit to a working demo that passes the above checks.