README.md · Agents-MCP-Hackathon/data_science

metadata

title: Data Science Agent
emoji: 🐢
colorFrom: green
colorTo: red
sdk: gradio
app_file: app.py
license: mit
sdk_version: 5.33.0
short_description: Agent that solves your AI/ML/Data Science problems
pinned: true
tags:
  - agent-demo-track
  - Mistral
  - LLamaIndex
  - Sambanova
  - Modal

Demo Video

🎥 Watch the Demo Video here

More details: Detailed Talk here

Acknowledgements

Made with ❤️ by Bhavish Pahwa & Abhinav Bhatnagar

Here’s the refined How It Works section with the iterative back‑and‑forth and LlamaIndex MCP integration clearly outlined:

🔧 How It Works

1. Gather Requirements

The user engages in a conversation with the chatbot, describing their data science / AI / ML problem.
There’s an iterative back-and-forth between the user and Gemini‑2.5‑Pro—the model asks clarifying questions, the user responds, and this continues until Gemini‑2.5‑Pro is satisfied that requirements are complete. Only then does it issue a “satisfied” response and release the structured requirements. ([huggingface.co][1], [youtube.com][2])

2. 🛠️ Generate Plan (button)

Clicking Generate Plan makes use of LlamaIndex’s MCP integration, which:
- Discovers all available tools listed via MCP on the Hugging Face server (hf.co/mcp) ([medium.com][3])
- Prompts Gemini‑2.5‑Pro again to select the appropriate tools and construct the plan workflows and call syntax.
All logic for tool discovery, orchestration, and MCP communication is deployed as a Modal app.

3. 🚀 Generate Code (button)

When the user clicks Generate Code, the Mistral DevStral model (served via vLLM, OpenAI-compatible) generates runnable code matching the plan and selected tools. This model, and its integration, are hosted on Modal Labs.

4. ▶️ Execute Code (button)

The Execute Code button sends the generated script to a sandboxed environment in Modal Labs, where it’s securely run. Execution results and any outputs are then presented back to the user.

This workflow flows user ↔ requirements collection ↔ tool planning ↔ code generation ↔ secure execution—with each step backed by powerful LLMs (Gemini‑2.5‑Pro, Mistral DevStral), LlamaIndex + MCP, and Modal Labs deployment. Samabanova models with Cline are used as devtools / copilots.

License

This project is licensed under the MIT License – see the LICENSE file for details.

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference