Brigada LLM
One chat for internal AI agents. A router (bge-m3 embeddings + qwen3:8b) sends each question to a Policy Agent with permission-aware hybrid retrieval and citations, or a Quality Rubrics Agent. Runs on local models, so policy text never leaves company servers. Evaluated on 90 policy questions and 76 routing cases.
Stack: Python, FastAPI, Ollama, qwen3:8b, bge-m3, pgvector, Supabase
Case study
Delai, AI Travel Companion
A travel assistant with 25 live tools (flights, hotels, tours, weather, tides, safety) and retrieval over 217 curated entries in pgvector. Answers stream as cards onto a live map, with guardrails against invented inventory and a live-model evaluation harness. Innovation Award, Top Executive Congress 2025.
Stack: Python, FastAPI, Gemini via OpenRouter, pgvector, Supabase, React
Case study · Live site
Brigadahub + Range AI
Booking and payments platform for a 30-bay golf driving range, pickleball courts and a kids playground, with a five-step booking flow, live bay availability and PayMongo checkout. Range AI answers guests on the website and Facebook Messenger and takes them from a question to a paid booking; accuracy on real guest messages rose from 55% to 71%.
Stack: React 19, TypeScript, Express 5, Supabase, PayMongo, Python, scikit-learn, Gemini, Messenger API
Case study · Live site