Statistical Analysis Agent Service

A FastAPI-based service that accepts a natural-language prompt describing a statistical analysis to perform on a configured CSV dataset, executes the analysis, and writes a structured result into a SQLite database table.

Features

• Natural language prompt to analysis mapping (basic patterns + safe eval of expressions)
• Pandas-based statistics and plotting (summaries, correlations, regressions)
• Results persisted to SQLite results table
• Simple REST API

Quickstart

1. Create a virtual environment and install dependencies:

   uv venv && source .venv/bin/activate || python -m venv .venv && source .venv/bin/activate
   pip install -r requirements.txt

2. Place your dataset CSV at data/dataset.csv (or set DATASET_PATH).
3. Run the server:

   uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

API

• POST /analyze:

  • Body:

    { "prompt": "compute mean of sepal_length grouped by species" }

  • Response: { "result_id": 1 }

• GET /results/{id}: fetch a single result row

• GET /results: list recent results

Configuration

• DATASET_PATH: path to CSV (default data/dataset.csv)
• DATABASE_URL: SQLite URL (default sqlite:///data/results.db)

Notes

• The NL-to-analysis layer uses simple pattern/rule-based parsing with guarded evaluation. For complex use, extend app/agent.py.