Google Summer of Code: Production RAG Application

[franciscojavierarceo]

Is your feature request related to a problem? Please describe.
We want to build a production service that empowers data scientists and machine learning engineers to use RAG through their feature store.

Describe the solution you'd like
Supporting NLP, LLMs, and RAG as a first class citizen in Feast requires enhancing Feast across several key areas.

1. We will need to create an explicit add on Introduce Feast NLP/LLM Add-On #4964
2. We will need to support tokenization libraries out of the box
3. We will need to be able to return different matrix/tensor types for the online response Return different matrix types for online serving #4714
4. We will need to be able to declaratively configure different types of search algorithms (e.g., using Milvus Add support for Milvus #4364) in FeatureViews
5. We will need to natively support a suite of transformations into context
   • For example, PDF/JSON/Markdown (via Docling)
6. We will need to support historical transformation of tokens into standard search algorithms (e.g., TF-IDF, BM25, bag-of-words, etc.) so that these algorithms / features can be used to fine tune the online hybrid search
7. We could support a nice UI so that users could see how a document is chunked (maybe in a Streamlit application)

Describe alternatives you've considered
N/A

Additional context
N/A

[HaoXuAI]

This is a really good feature. I think it would be really useful to develop a toolkit, maybe something called feast-llm, which can be used by end users to create feature transformation without touching libraries like torch or sklearn.

[franciscojavierarceo]

FYI @andreyvelich

Feast was not selected for GSoC this round. Maybe next year.