yahskapar/personal-health-insights-agent

Processes wearable device data (steps, heart rate, sleep) through an LLM agent that answers both structured queries ("longest run distance") and open-ended health questions ("how to reduce stress") by reasoning over temporal patterns. Built on Google's Gemini API with few-shot prompting and integrated web search via Tavily for supplemental health guidance. Includes evaluation datasets with 4,000+ objective queries and 172 open-ended queries across real and synthetic user profiles with deidentified health metrics.