Skip to content

Project Ideas

Ivan Ogasawara edited this page Feb 11, 2025 · 4 revisions

RAGO

Project Idea 1: Expand Backend Support for Augmented and Generation Modules

Abstract

Rago is a lightweight framework for RAG. Its main idea is to provide a RAG with a very beginner friendly structure for anyone who wants to start to work with this concept.

Current State

RAGO currently supports OpenAI as the backend for both the Augmented (Embeddings) and Generation (LLM) modules. To increase flexibility and interoperability, this project aims to add support for multiple additional backends across both modules. This requires prior investigation to determine if each provider offers a Python API that can be integrated into RAGO.

Tasks

  • Investigate the availability and API support of the following embedding providers:
    • AI21
    • Ollama
    • Together
    • Fireworks
    • MistralAI
    • Cohere
    • Nomic
    • Databricks
    • VoyageAI
    • IBM
    • NVIDIA
  • Investigate the availability and API support of the following LLM providers:
    • AI21
    • Anthropic Claude
    • Bedrock
    • Cohere
    • DeepSeek
    • Fireworks
    • Ollama
    • Together
    • Vertex
    • NVIDIA
  • Implement integration for supported backends.
  • Ensure compatibility with RAGO's existing framework.
  • Write tests and documentation for each newly integrated backend.

Expected Outcomes

  • RAGO supports multiple embedding and LLM backends.
  • Clear documentation and examples for integrating and using different providers.
  • A robust test suite to ensure stability across different backends.

Details

  • Prerequisites:
    • Experience with Python and API integrations (intermediate level)
    • Familiarity with LangChain or similar frameworks (beginner level)
    • Understanding of Retrieval-Augmented Generation (RAG) workflows (beginner level)
    • Knowledge of embedding models and LLM APIs (beginner level)
  • Duration: 350 hours
  • Complexity: Medium
  • Potential Mentor(s): Ivan Ogasawara, Ever Vino, Sandro Loch, Alexandre de Siqueira

References