Summary
AgentRAG currently documents the intended modular retrieval architecture well, but the repo is still README-only. The best first implementation milestone is a thin set of interfaces that define the retrieval pipeline shape.
Why this matters
Without a clear interface layer, later work on chunkers, embedders, retrievers, rerankers, and vector stores will drift quickly. A small protocol/abstraction layer would make the project direction concrete.
Proposed scope
- define the first core interfaces or protocols for chunking, embedding, retrieval, and reranking
- add a minimal pipeline composition flow that wires those stages together
- include one in-memory/example implementation for development
- document the intended stage boundaries in code and README
Acceptance criteria
- the repo contains a small but usable code skeleton for the retrieval pipeline
- interfaces are clear enough for later provider/back-end implementations
- one example demonstrates indexing + retrieval flow end to end
- README reflects the first implemented architecture
Notes
The goal here is not feature completeness. It is to establish a stable, modular starting point for the rest of the project.
Summary
AgentRAG currently documents the intended modular retrieval architecture well, but the repo is still README-only. The best first implementation milestone is a thin set of interfaces that define the retrieval pipeline shape.
Why this matters
Without a clear interface layer, later work on chunkers, embedders, retrievers, rerankers, and vector stores will drift quickly. A small protocol/abstraction layer would make the project direction concrete.
Proposed scope
Acceptance criteria
Notes
The goal here is not feature completeness. It is to establish a stable, modular starting point for the rest of the project.