Autonomous Data Analysis

An agent-based system that turns structured datasets into clear, verified insights. LangGraph coordinates planning, preprocessing, analysis, interpretation, review, and presentation.

Architecture

flowchart LR
    D[(CSV / Database)] --> P[Planner]
    P --> R[Preprocessing]
    R --> A[Analysis]
    A --> I[Interpretation]
    I --> C{Critic}
    C -->|Approved| H[Chat]
    C -->|Revise| P
    H --> U[User / App]

Loading

The agents share validated tools for dataset inspection, preprocessing, statistics, visualization, and optional machine learning.

Prerequisites

• Git
• Python 3.12.13
• uv 0.11.21

Setup

git clone https://github.com/CogitoNTNU/autonomous-data-analysis.git
cd autonomous-data-analysis
uv sync --project backend
uv run --project backend pre-commit install --config .config/pre-commit-config.yaml

Usage

Inspect the included penguins dataset:

uv run --project backend python -m backend.app.tools.inspect_dataset

The command prints the dataset dimensions, missing values, inferred datatypes, and duplicate-row count.

To inspect another CSV file from Python:

from pathlib import Path
from backend.app.tools.inspect_dataset import inspect_dataset

inspect_dataset(Path("path/to/dataset.csv"))

Project structure

.
├── backend/
│   ├── app/       # API, agents, and analysis tools
│   ├── data/      # Example datasets
│   ├── tests/     # Backend tests
│   ├── pyproject.toml
│   └── uv.lock
├── frontend/      # React application
├── docs/          # Documentation content and MkDocs config
├── .config/       # Shared development-tool configuration
├── .python-version # Shared Python version for local tooling and uv
└── README.md

Testing

uv run --project backend pytest backend/tests

Team

Kien Parajes Le

Sebastian Jøranger

Kristoffer

Daniel Samoylov

License

MIT