Homestagr

Most real estate agents waste hours bouncing between photo editors, listing portals, and copy templates. Homestagr collapses all of that into one WhatsApp chat — send the photos, get enhanced images, platform-specific listings, fly-through videos back, and also surfaces up-to-date market data for the areas they cover.

The goal was one conversation where they send photos and get back everything ready to publish. That “one conversation” constraint pushed me toward a single-agent architecture with tool-calling — one LLM (OpenAI Responses API) that orchestrates tools, routing between Gemini for image enhancement, Replicate/Kling for video generation, and Supabase for storage and state.

The hardest part was WhatsApp. Messages arrive as separate webhooks with no grouping, so I built a batch accumulator in Redis with settling logic — the system waits a few seconds until photos stop arriving, drains them atomically, classifies each to reject non-property images, and only then spends credits on enhancement.

The stack

I built it on FastAPI backend with async workers (ARQ + Redis), Next.js frontend, Supabase (auth, database, storage), Stripe for credit-based monetization, and WhatsApp Cloud API integration. Deployed on DigitalOcean with GitHub Actions CI/CD.

The brand

The brand system below is part of my work as well — logo variations, icon set, tagline, ad concepts, social templates, and a navy palette tuned to read as modern, trusted, and premium across every surface a listing touches.