Chatbot showing results
Chatbot available
Filling in Information for evaluation
Main Page
Impact Score Card
Available accommodations in respected areas
Searching near Chula
Evaluate a lease
Chatbot activated
Link towards google map
Available accommodations in respected areas
Inspect Safety Score Card
Main Page: Dark Mode

Inspiration

Renting an accommodation in Bangkok often feels like navigating a minefield of non-transparent lease terms. Renters, especially university students and relocators, frequently encounter hidden fees, illegal move-in deposits exceeding standard limits, and inflated private sub-meter electricity rates that charge up to triple the official Metropolitan Electricity Authority (MEA) tariff.

We built SafeRoom to establish complete transparency in Bangkok rental living, pairing strict local tenancy math with real geographic accommodation datasets and an intelligent assistant powered by Google Gemini.

What it does

Multi-District Verified Housing Directory: Catalogs accommodations across Bangkok's administrative districts with real transit references and direct deep links to official Google Maps coordinates.
Strict Tenancy Compliance Math: Enforces local Thai rental laws automatically:
Condominiums: Audits standard (2 + 1) month deposits (2 months security + 1 month advance) and verifies direct MEA government meter electricity (~4.5 THB/unit).
- Apartments & Dormitories: Audits standard (1 + 1) month deposits and highlights private sub-meter surcharges (7.0–8.0 THB/unit).
Lease Evaluation Engine: Allows users to input contract terms (rent, advance payment, deposit, utility surcharges) and computes an instant transparent risk score and fairness verdict.
Rental Assistant (Powered by Google Gemini 2.5 Flash): An AI conversational assistant with built-in intent parsing and phonetic typo tolerance. Renters can search naturally (e.g., "Find rooms near MRT Bang Wa under 10k" or misspelled queries like "Kha Na Yo under 8,000"), and Gemini returns verified properties strictly isolated to that district with clickable Google Maps links, clean formatting without markdown asterisks, dynamic context chips, and a one-click chat reset.
Daylight & Night Mode UI: Tailored interface with responsive toggles for seamless usability in any environment.

How we built it

Frontend & Styling: Built using React 18, Vite, and Tailwind CSS with custom daylight-first and Night-mode themes.
AI Core: Integrated Google Gemini 3.5 Flash via the Google AI Studio API (``REST endpoint) grounded with localized district metadata and tenancy rules.
Intent & Fuzzy Search Engine: Implemented client-side Levenshtein distance matching across Bangkok administrative districts, subdistricts (khwaeng), transit hubs (BTS, MRT, ARL, SRT), and price constraints.
Deployment: Deployed on Vercel with automatic single-page application (SPA) rewrites.

Challenges we ran into

District Intent Bleed: Early iterations sometimes defaulted to fallback properties in other districts when parsing complex or misspelled queries. We resolved this by building a dedicated 50-district dictionary and enforcing strict query isolation before Gemini formats the response.
Output Cleanliness: Prompting the model to omit raw markdown asterisks while maintaining readable typography required strict system instructions and client-side sanitization.

Accomplishments that we're proud of

Delivering a verified housing platform with direct Google Maps navigation for every card.
Building an AI assistant that handles informal student queries and budget filters accurately.
Achieving a clean deployment on Vercel with zero runtime errors and fast client response times.

What we learned

How to ground generative models with structured JSON datasets to eliminate hallucinations in real-estate pricing.
The power of combining deterministic fuzzy logic with LLM natural language understanding.

What's next for SafeRoom

Adding multi-language contract PDF scanning via Gemini Vision to audit physical lease agreements directly from photos.
Expanding verified coverage deeper into fringe residential corridors.