Steering Gemini for well being teaching
“Do I get higher sleep after exercising?” seems like a easy query, however answering it like a proactive, personalised and adaptive coach required a number of technical improvements.
First, we’d like the coach to grasp and do numerical reasoning on physiological time sequence information resembling sleep and exercise, utilizing capabilities much like these showcased by PH-LLM. For questions like this, the coach verifies latest information availability, chooses the suitable metrics, contrasts related days, contextualizes outcomes in opposition to private baselines and population-level statistics, incorporates prior interactions with the coach, and eventually makes use of the evaluation to offer tailor-made solutions and insights.
Second, we make the most of a multi-agent framework that coordinates skilled sub-agents to offer clear, constant and holistic assist, resembling (1) a conversational agent for multi-turn conversations, intent understanding, agent orchestration, context gathering and response era; (2) a knowledge science agent that iteratively makes use of instruments to fetch, analyze, and summarize related information (e.g., sleep and exercise information), leveraging code-generation capabilities as wanted; and (3) a website skilled, resembling a health skilled that analyzes consumer information to generate personalised health plans and adapt them as progress and context change.
