Hands-on Python experiments for RAG, embeddings, vector search, and evaluation
Awesome LLM Experiments is a Python repo of focused walkthroughs—Jupyter notebooks today, with room for plain Python modules as experiments grow—covering retrieval-augmented generation, embeddings, Chroma vector search, LangChain wiring, multilingual and multimodal retrieval, and RAGAS evaluation. Each experiment lives in its own folder.
When it is useful
You are learning or teaching how modern RAG pipelines fit together; you want small, copy-friendly examples instead of a full product codebase; or you need a starting point before building production retrieval for a real app.
What you can do
- Run multilingual RAG on bilingual UDHR text (English + Hindi) with E5 embeddings, Chroma, optional language filters, and cross-lingual search.
- Open the multimodal text+image notebook in Colab with a GPU for cross-modal VL embeddings (Qwen3-VL via Sentence Transformers).
- Evaluate an English RAG pipeline with RAGAS metrics for retrieval (context precision/recall) and generation (faithfulness, answer relevancy).
- Install dependencies with
uv syncand run each experiment from the repo root or its folder—notebook or script, as that experiment documents.
Limits
- These are learning experiments, not production services—no auth, scaling, or deployment story baked in.
- The multimodal notebook is Colab-only and not installed via
uv sync. - RAGAS and embedding runs may need API keys and compute; results depend on models and sample data, not universal benchmarks.
Setup, experiment table, and paths are in the GitHub repository.
