4sight's AI assistant is called Galen, named after the ancient physician who pioneered the idea that health is a balance of interconnected systems. Galen uses a technique called Retrieval-Augmented Generation (RAG) to analyze your personal health data and answer questions in plain language. It is a Premium feature available with a free trial.
What Is Retrieval-Augmented Generation?
RAG is an AI architecture that combines two capabilities: information retrieval and language generation. Instead of relying solely on a general-purpose language model that was trained on public data, Galen first retrieves the specific health data relevant to your question and then generates a personalized answer grounded in that data.
The process works in three steps:
- Vector embeddings: Your health data is converted into numerical representations (vector embeddings) that capture the meaning and relationships between different metrics. These embeddings are stored securely and updated as new data arrives.
- Semantic search: When you ask a question, Galen converts your question into a vector and searches your personal health data to find the most relevant information. This is not a simple keyword search — it understands the intent behind your question.
- Language model generation: The retrieved data is sent as context to a large language model, which generates a natural-language response that directly answers your question based on your actual health data.
This means Galen never hallucinates answers about your health. Every response is grounded in data that was actually retrieved from your personal health timeline.
The 9 Query Types
Galen is not a generic chatbot. It is specifically designed to understand health data queries and routes your questions to the appropriate analysis pipeline. There are nine supported query types:
- Lookup — Retrieve a specific value. "What was my weight yesterday?"
- Trend — Analyze how a metric changes over time. "How is my resting heart rate trending this month?"
- Comparison — Compare time periods or conditions. "How does this week compare to last week for sleep?"
- Correlation — Find relationships between metrics. "What affects my sleep quality the most?"
- Dose-response — Quantify how one variable affects another. "How does caffeine affect my sleep onset time?"
- Journey — Trace your progression toward a goal. "How did I get from 280 to 260 pounds?"
- Top-N — Find your best or worst entries. "What do my 10 best sleep nights have in common?"
- Aggregation — Calculate averages, totals, and summaries. "What's my average step count on weekdays?"
- Conversational — General health questions and follow-ups. "What should I focus on to improve my score?"
When you ask a question, Galen's query classifier automatically determines which type of analysis to perform. You do not need to use any special syntax — just ask in plain language.
Nightly Correlation Engine
Beyond answering questions on demand, Galen runs a nightly correlation computation across your entire data set. This background process analyzes relationships between metrics from all four pillars and precomputes correlation coefficients, statistical significance, and effect sizes.
The nightly engine is what powers two important features:
- AI Insights — Precomputed findings that appear on your dashboard, such as "Your mood scores 34% higher on days you sleep 7+ hours" or "Your deep sleep improves after days with 8,000+ steps."
- Proactive notifications — When the system detects a new meaningful pattern or a significant change in an existing pattern, it sends you a notification. You do not have to ask the right question — Galen surfaces the insight for you.
This is one of the key differences between Galen and a general-purpose AI chatbot. Galen is continuously analyzing your data in the background, not just waiting for you to ask questions.
Safety Guardrails
Health data analysis carries responsibility. Galen includes multiple layers of safety guardrails to ensure it provides helpful information without overstepping into medical advice:
- Emergency detection: If your question or data suggests an emergency situation (e.g., chest pain, suicidal ideation), Galen immediately directs you to call emergency services or a crisis hotline. It does not attempt to provide medical advice in emergencies.
- Medical disclaimer: Every response from Galen includes a reminder that its insights are for informational purposes only and are not a substitute for professional medical advice, diagnosis, or treatment.
- Scope boundaries: Galen will not diagnose conditions, prescribe treatments, or make claims about medical causation. It reports correlations and patterns in your data and encourages you to discuss findings with your healthcare provider.
- Data grounding: Because Galen uses RAG, it only generates responses based on your actual data. It does not speculate about data it does not have.
What You Can Ask
Here are examples of real questions you can ask Galen across all four pillars:
- "What was my average HRV this week versus last week?"
- "Show me my sleep trend for the last 30 days."
- "What do my best mood days have in common?"
- "How does my protein intake correlate with my workout performance?"
- "When is my caffeine usually cleared before bed?"
- "How has my weight changed since January?"
Galen works best when you have data from multiple pillars, because cross-pillar correlations are where the most valuable insights emerge. The more data you give 4sight, the smarter Galen becomes about your personal health patterns.