Skip to content

Rag build specifically for Polish language. By default it uses Bielik 11b-8bit. All of those components (llm, embeddings model and reranker) are meant to be self hosted. It is meant to be self hosted

License

Notifications You must be signed in to change notification settings

Kurdzik/polish-rag

Repository files navigation

Local RAG PL

Project that enables for local deployment of full rag backend that performs especially well in Polish language


Tech stack

All services are hosted with openai-compatibile api therefore we can connect to them using unified connectors

  • Base llm - Bielik 11b 8bit
  • Embeding model - snowflake-arctic-embed-l-v2.0
  • Reranker - sdadas/polish-reranker-bge-v2

Databases

  • Vector database - Qdrant
  • Fuzzy search db - Elasticsearch
  • App database - Postgres

Default tracer - Opik


TODO

  • Implement Active Directory login
  • Implement Auth middleware and JWT generation
  • Implement Basic chat view as a native component of the app

About

Rag build specifically for Polish language. By default it uses Bielik 11b-8bit. All of those components (llm, embeddings model and reranker) are meant to be self hosted. It is meant to be self hosted

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published