# autoresearch

**Repository Path**: yuxx1006/autoresearch

## Basic Information

- **Project Name**: autoresearch
- **Description**: AI agents running research on single-GPU nanochat training automatically
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-03-20
- **Last Updated**: 2026-03-20

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# AutoHealth - Personal Health Data Analysis System

![health](health-icon.png)

*An autonomous health analysis system that learns from your personal health data and provides increasingly personalized recommendations over time.*

## Concept

Similar to how autoresearch autonomously iterates on ML model training, AutoHealth autonomously analyzes your health data and iterates on recommendations. Each time you provide new health data (biomarkers, nutrition, exercise, sleep), the system:

1. **Ingests** the new data into your personal health database
2. **Correlates** the data to find patterns (e.g., CRP ↑ correlates with sleep ↓)
3. **Generates** personalized recommendations based on correlations
4. **Learns** from feedback to improve future recommendations
5. **Tracks** recommendation effectiveness over time

The system runs automatically via GitHub Actions once per day, analyzing all available data and updating recommendations.

## How It Works

The repo has three core files:

- **`prepare.py`** — Fixed utilities for data ingestion, parsing (PDFs, images, CSVs), and health metric calculations. Not modified.
- **`analyze.py`** — The analysis engine that the agent iterates on. Contains correlation detection, recommendation generation, and learning logic. **This file is edited by the agent**.
- **`program.md`** — Instructions for the health analysis agent. **This file is edited by the human**.

## Quick Start

**Requirements:** Python 3.10+, [uv](https://docs.astral.sh/uv/)

```bash
# 1. Install dependencies
uv sync

# 2. Initialize health database
uv run prepare.py

# 3. Add your first health record (example)
uv run prepare.py --add-sample

# 4. Run analysis manually
uv run analyze.py
```

## Data Types Supported

- **Biomarkers**: Blood tests, CRP, glucose, cholesterol, etc. (PDF, images, CSV)
- **Nutrition**: Food logs, meal photos, macros (images, JSON, CSV)
- **Exercise**: Workouts, steps, heart rate (JSON, CSV, API integrations)
- **Sleep**: Duration, quality, stages (JSON, CSV, API integrations)
- **Subjective**: Energy levels, mood, symptoms (text, ratings)

## Project Structure

```
prepare.py           — data ingestion, parsing utilities (do not modify)
analyze.py           — correlation engine, recommendations (agent modifies)
program.md           — agent instructions
data/                — your personal health records
  ├── biomarkers/    — blood tests, lab results
  ├── nutrition/     — food logs, meals
  ├── exercise/      — workouts, activity
  ├── sleep/         — sleep data
  └── subjective/    — mood, energy, symptoms
recommendations/     — generated recommendations
  ├── daily/         — daily analysis results
  └── insights/      — long-term patterns
.github/workflows/   — daily automation
```

## Running the Agent

Point your Claude agent to `program.md`:

```
Hi have a look at program.md and let's kick off a new health analysis cycle!
```

## GitHub Actions Automation

The system runs automatically once per day via GitHub Actions:

1. Pulls latest health data from `data/`
2. Runs analysis with current `analyze.py`
3. Generates daily recommendations
4. Commits results back to repo
5. Creates issues for important health insights

To enable:
1. Push this repo to GitHub
2. Add your health data files to `data/` directories
3. GitHub Actions will run daily at 6 AM UTC

## Privacy

All health data stays local in your repo. Only you have access. GitHub Actions runs in your private repo.

## License

MIT