05 · Multi-agent · Hackathon
Sandy AI Lab
An overnight multi-agent lab assistant — top 5 at Code It Up 6.0.
Ask the lab a question in plain language, typed or spoken. A planner coordinates an inventory agent and a research agent over live data, and a Groq-hosted LLM answers — constrained to the context it is given.
What I built
- 01
Three agents — inventory (minimum stock), research (blocked or delayed projects) and a planner that turns anomalies into a prioritised action list.
- 02
A three-tier data fallback — external REST API, then local PostgreSQL, then a static structure — so the demo kept answering when upstream services failed.
- 03
A streaming chat interface with voice input and spoken replies through the Web Speech API.
Evidence
