Feature Request: Integrate ragbio for Knowledge-Aware Cluster Summarization

[man4ish]

We maintain two separate projects: scatlas-builder, the core single-cell analysis platform, and rag-gene-discovery-assistant (ragbio), the RAG-LLM–based biological knowledge extraction tool.
The objective is to integrate ragbio into scatlas-builder to generate biologically grounded, citation-rich summaries for cell clusters.

This integration is a key step toward evolving scatlas-builder into a full scientific discovery platform that combines computational analysis with literature-backed biological interpretation.

---

Acceptance Criteria

1. New API Endpoint

Create a new FastAPI endpoint to provide LLM-augmented summaries for cluster marker genes.

• Path: /summarize/cluster/{dataset_id}

• Method: POST

• Input:
  JSON containing the dataset_id and a dictionary of cluster markers, e.g.:

  {
    "Cluster 0": ["GeneA", "GeneB", "GeneC"],
    "Cluster 1": ["GeneD", "GeneE"]
  }

• Output:
  JSON with biological interpretations and PubMed-derived citations for each cluster.

---

2. Service Layer Integration

The new endpoint must call a dedicated service function—for example:
llm_service.run_rag_summary()—that invokes the ragbio RAG pipeline and returns structured summaries.

---

3. Deployment Updates

Modify:

• docker/Dockerfile
• requirements.txt

to ensure that ragbio and its dependencies (Ollama runtime, FAISS index, BioBERT embeddings, etc.) are available within the scatlas-builder deployment environment.
Alternatively, configure scatlas-builder to communicate with an external Ollama/RAG service.

---

4. Diagram + Architectural Explanation

Provide a clear architectural explanation and a System Diagram showing how:

• client requests flow through the new FastAPI route,
• marker genes reach the analysis layer (Scanpy),
• results pass into the LLM orchestration layer,
• ragbio performs retrieval and LLM reasoning,
• summaries return to the client.

---

System Integration Diagram (Reference Implementation)

This diagram illustrates the expected data and control flow needed to integrate the ragbio knowledge layer into the scatlas-builder analysis layer.