Simple RAG as a service
This a sample WIP rag project.
This RAG code interfaces with a Cosmos DB setup in azure. The Cosmos DB is used to store the vector data of embeddings.
This project has a simple vueue.js front end project. The front end allows the user to upload a PDF embeddings to cosmos DB. The front also allows the user to ask questions in regards to the PDF against an LLM.
The processing is performed using a Rest service implemented using an azure function. No llm framework such as lang chain is used in this case.
The project works as follows.
Upload of File
-User uploads file. -File is chunked and submitted to LLM to retrieve embeddings. -Embeddings are store in a cosmos DB DB in azure.
Chat against PDF.
-The system prompt is setup such that the embeddings that are passed are used to determine the answer.
- The user asks a question.
- The question is used to extract relevant embeddings using cartesian lookup.
- The embeddings are passed along with the question to the LLM (in this case chatgpt).
- The LLM then provides an answer using the embeddings.
Enhancement in progress.
-Google auth to allow multiple user to upload. Currently the db does have a user field but it is unused. -Allow user to select PDF files searched. Also save PDF file description. -Hybrid search and multi stage pipeline using graph and text search. Currently this design has the limitations of a normal vector rag
Github of current source :