Akhra

Akhra is our Smart India Hackathon prototype for problem statement 26043, set by the Government of Jharkhand. The brief asks for a platform that crowdsources societal problems and routes them to universities and industry. We cleared the college's internal round and submitted to the SIH portal. I did the development, design and architecture; five teammates handled research, testing, UI feedback and the presentation.

The first thing I changed was the brief's own shape. Most of what people actually report is ordinary: a chapakal that stopped working, a blocked drain. Routing that to a research team is theatre. So an officer chooses one of two tracks. Routine problems go to the line department with a 21-day deadline. Problems bigger than one repair become university projects with proposals, milestones and field tests, with industry partners offering funding, mentoring or pilots. Either way the reporter follows one timeline, confirms the fix, and can reopen it within 30 days.

A report can't be allowed to fail because an AI service is busy, so classification is a chain: Gemini, then Groq, then a TF-IDF model that runs on our own server. Duplicate detection had the same problem across languages. "Chapakal kharab hai" and "handpump broken" share no words, so a small lexicon folds Hindi, Devanagari and Hinglish spellings onto the same concepts before comparing, and the dashboard shows which tier actually answered.

District officers see only their own district's reports. That's enforced by PostgreSQL row-level security as well as the application, so a leak needs two mistakes rather than one forgotten where-clause. We also had a rule: nothing that needs money or a registered company. SMS was removed entirely rather than stubbed, because Indian transactional SMS requires a DLT-registered legal entity, and a feature that can't be shown working shouldn't pretend to exist.