Agentic Data Sampler — FastAPI SaaS Prototype

A modular, token-aware data sampler that uses an LLM agent to decide structured & unstructured sampling strategies. This repo demonstrates turning the sampler into a FastAPI SaaS app.

Features

• Structured (Parquet/CSV) profiling via DuckDB
• Unstructured text chunking + optional semantic sampling (sentence-transformers)
• Pluggable agent adapter (Bedrock/OpenAI/local)
• Token estimation, budget enforcement, and fallbacks
• Async FastAPI endpoints + background tasks

Getting started (dev)

1. Copy environment variables into .env.

2. Build & run: docker build -t sampler-app . docker run -p 8000:8000 --env-file .env sampler-app

3. Open docs: http://localhost:8000/docs

Recommended next steps

• Add persistent storage (Postgres), API key auth, and Prometheus metrics.
• Add unit & integration tests.

License: MIT