NexusRAG

One tested slice a day

NexusRAG is a corrective, multimodal RAG system built in thirty daily slices.

Read the code See the thirty days
12
of 30 days built and tested

Five stages from query to citation

  1. Router

    Picks the retrievers for each query.

    Planned

  2. Hybrid retrieval

    Qdrant dense search beside SQLite keyword search.

    Built

  3. Fusion and reranking

    Reciprocal Rank Fusion, then a cross-encoder.

    Built

  4. Correction

    Grades evidence sentence by sentence.

    Planned

  5. Generation and grounding

    Answers from surviving evidence, with citations.

    Planned

Pick a day, see its slice

    Day 12, cross-encoder reranking

    Core RAG foundation. Built and tested.

    Every new feature must have a provider boundary, tests, and an evaluation path.

    Before you clone it

    What runs today?
    Document loaders, chunking, embeddings, semantic and keyword retrieval, fusion and reranking.
    What is not built yet?
    The agents, corrective grading, generation, multimodal search, memory and evaluation.
    What does it need?
    Python 3.11 or later. Docker Compose starts Qdrant and Redis.

    Run it on your machine

    FastAPI serves its own docs at localhost:8000/docs.

    Open the repository 
    git clone https://github.com/boss477/NexusRAG.git
cd NexusRAG
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env
uvicorn app.main:app --reload
    On Windows, activate with .venv\Scripts\activate and use copy in place of cp.