feat: RAG pipeline with Qdrant and sentence-transformers on Crusoe Managed Inference

[Sakshi3027]

What this adds

A full RAG (Retrieval-Augmented Generation) pipeline running on Crusoe Managed Inference.

Pipeline:
Documents → Embeddings → Qdrant → Retrieve top-k → Generate grounded answer

How it works

1. Embeds documents using all-MiniLM-L6-v2 (sentence-transformers, 384-dim)
2. Stores vectors in Qdrant in-memory (zero infrastructure setup)
3. At query time, retrieves the top-3 most semantically relevant chunks
4. Passes retrieved context to Crusoe Managed Inference to generate a grounded answer

Why it's useful

RAG is one of the most common patterns for building LLM applications. This gives teams a working, runnable starting point for building RAG on Crusoe no external services required out of the box.

The vector store is swappable in one line: in-memory → local persistent → Qdrant Cloud.

Testing

Tested locally using Groq as a drop-in replacement. All 3 queries returned accurate, context-grounded answers.

To run on Crusoe:
export CRUSOE_API_KEY="your-api-key"
python example.py

Related contributions

• feat: LangGraph multi-node research agent on Crusoe Managed Inference #55 — LangGraph multi-node agent on Crusoe
• feat: MLflow experiment tracking for Crusoe Managed Inference #56 — MLflow experiment tracking on Crusoe