SetScout

LangGraph · Hugging Face · Kaggle

Problem

Finding the right public dataset means tab-hopping across Hugging Face, skimming READMEs, and mentally scoring fit against your constraints. SetScout automates that loop: describe what you need in plain language, and a four-node LangGraph pipeline searches sources, fetches evidence, and returns a structured markdown report with per-dataset requirement checks and rankings.

Pipeline

Decomposer. Turns form inputs into a SearchSpec with keywords, MeSH terms, sources, and hard constraints. LLM with rule-based fallback.

Searcher. Parallel async search across Hugging Face and Kaggle. Returns up to 8 candidates.

Gather evidence. Fetches dataset cards and README excerpts in parallel.

Evaluator. Single batch LLM call scores all candidates: requirement checks, known issues, fit summaries, and final ranking.

Stack

LangGraph, LangChain, Pydantic, Gemini API. Optional Langfuse tracing and Kaggle credentials for extended search.

Status

Active development. A Gradio UI and Hugging Face Spaces deployment are in progress; live demo coming soon.