janikdotzel/healthcare-agent

Personal Health Advisor is an AI assistant that brings together your health data from multiple sources, fitness trackers like Fitbit, medical records, and other sensors, into one chat interface. Instead of jumping between apps to check your steps, sleep, heart rate, and doctor's notes, you ask questions in plain language and get answers that draw on all of that data at once. When you ask a question, the agent figures out which data sources it needs to check. It might pull your recent heart rate from sensor data stored in its database, fetch yesterday's steps and sleep from the Fitbit API, and search your medical records for relevant context. It then combines all of that information and passes it to a language model that writes you a natural-language answer. The system also accepts incoming health data through a separate ingestion endpoint, so sensor readings and medical records can be fed in continuously and queried later. A person might use this if they want a single conversational interface for their health. For example, you could ask why you last visited the doctor, and the agent would pull up your diagnosis and prescribed medication. You could ask how much REM sleep you got each day last week, and it would fetch that from Fitbit. You could even get combined insights, like a reminder to exercise more because your medical notes mention back pain from sitting all day, and your step count has been low. One notable design choice is how medical records are handled. Rather than stuffing entire documents into the AI model, the system uses a retrieval approach where it searches for only the relevant portions of your medical history based on your question. The project is built in Java using the Akka toolkit, and the README mentions plans for durable workflow orchestration to make the agent more resilient to failures, though that isn't fully implemented yet.