Homie — "Should I buy this flat?"

About eighty percent of Singaporeans live in HDB flats, and the resale market is where lease decay, school catchments, MRT distance and Master Plan zoning all collide. Today that research takes hours across a dozen government portals, Google Map searches, school location finding, Excel spreadsheets & Google Docs for calculating + combining findings, and social media sites such as Reddit for opinions. Homie answers the question in two minutes. Paste a homes.hdb.gov.sg resale URL and Homie, a multimodal Gemini agent built on Vertex AI, fans out across eight official Singapore data sources in parallel: HDB's resale portal, data.gov.sg HDB Resale Transactions, OneMap, LTA DataMall, URA Planning Decisions, the MOE School Directory, NParks, and the listing photos themselves analysed with Gemini Vision. A synthesis pass scores eight verdict dimensions tuned for HDB resale concerns on a 0 to 10 scale and assigns a Buy, Negotiate or Skip tag with a tagline and supporting paragraph. What makes Homie distinctive is the loop. The reasoning is streamed live as the agent dispatches tools. Bridging thoughts between calls make it feel like watching an analyst think. Once the verdict lands, a conversational layer lets you ask grounded follow-ups ("how does this compare to the block average and what can I use to negotiate further?") against the tool results, with a strict system prompt that refuses out-of-scope or speculative answers. Stack: Next.js 16, FastAPI, Gemini 2.5 Flash + Pro + Vision with function-calling, Playwright. Deployed on Vultr Cloud Compute for the AI Agent Olympics.