A multi-modal AI-powered property inspection platform built to turn field photos and voice notes into structured inspection reports, with a focus on automated issue detection, severity scoring, and cost estimation.

I implemented an inference pipeline on HuggingFace Inference Providers that runs uploaded photos and voice notes through object detection, severity classification, and speech-to-text models, converting raw field media into structured findings without manual write-up.

On the pattern-matching side, I built vector similarity search over pgvector to match new findings against a history of past inspections, surfacing comparable issues and cost estimates automatically. Photo and voice uploads are processed asynchronously through a queue, with Supabase Realtime pushing status updates to the client as each finding completes.

• GitHubhttps://github.com/Manpreet-Bhatti/inspectai
• PlatformWeb
• StackNext.js, React, TypeScript, Supabase, Python, FastAPI, Turborepo