ICICLE AI Chatbook

An interactive marimo notebook that turns the ICICLE AI Tapis services into a hands-on RAG (retrieval-augmented generation) playground. Paste or upload a document (PDF, DOCX, TXT, MD, up to 2 MB) and chat against it — or come back later and chat against a collection you saved before. Every step is visible, from embeddings and chunking to retrieval, reranking, grounded prompting and LLM-as-judge evaluation, so newcomers to RAG can see how each piece works.

References

• ICICLE AI Tapis portal — where you log in and grab an access token.
• Embedding service (icicleaiembedserver) — Qwen3-Embedding behind a JWT-gated FastAPI.
• Vector service (icicleaivecserver) API docs — Qdrant-backed store with private per-user collections, cosine retrieval, and reranking (MMR, cosine rescore, cross-encoder).
• marimo documentation — the reactive Python notebook used to host this playground.
• uv documentation — the package/env manager used to bootstrap the project.

Acknowledgements

National Science Foundation (NSF) funded AI institute for Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE) (OAC 2112606)

Issue reporting

File bugs, ideas, or questions on the GitHub issues page.