What it does

Finding a basa (flat) in Dhaka means dealing with vague listings (“5 min from Metro” often means 2.5 km away), hidden utility/service charges, advance-payment traps, and landlords (Bariwalas) who immediately reject bachelors and university students. Basa Khuji AI helps both sides of the rental market: tenants can decode any listing or To-Let sign photo, and landlords can post their flats for free.

Key Features

• Reality Report: Paste a listing text or upload a To-Let sign photo. Gemma returns a Bachelor Survival Score, a verdict stamp, a “What they wrote vs. What it actually means in Dhaka” translation, and an itemized receipt of the real monthly cost including hidden extras.
• Banglish Phone Script: Generates a polite, respectful Banglish call script tailored for university students to break the ice.
• Bariwala Boss Fight: An interactive roleplay chat with a simulated grumpy Dhaka landlord. Polite, honest answers raise Uncle’s patience meter; careless answers drop it.
• Two-Sided Marketplace: Landlords can post flats (basa.db via SQLite) with rent, floor, area, and tenant preferences, allowing house-hunters to run any posted flat through the AI decoder in one click.
• Gemma-First with Automatic Fallback: Defaults to Google’s open gemma-3-27b-it model and automatically falls back to gemini-2.5-flash if rate-limited (or runs in zero-key demo mode offline).

How we built it

Built with Python (Flask), SQLite, and Vanilla JavaScript calling Gemma and Gemini via the Google AI Studio API. Because the Gemma endpoint does not take a separate system instruction field, app.py injects system_prompt.txt into the first user turn to drive a two-mode state machine: Turn 1 outputs structured Markdown that the frontend parses into visual score rings and cost receipts, while Turn 2+ transitions seamlessly into the Bariwala roleplay boss fight.