The system supports four Graph RAG backends — LightRAG, MiniRAG, PathRAG, and HyperGraphRAG — all switchable at runtime with no code changes required. Each backend uses a different graph construction model, retrieval strategy, and output format, making the choice of backend a meaningful architectural decision for your specific use case.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/avnlp/agentic-med-diag/llms.txt
Use this file to discover all available pages before exploring further.
Backend Comparison
| Backend | Graph Model | Retrieval Mode | Neo4j Required | Context Format |
|---|---|---|---|---|
| LightRAG | Dual-level KG (local entity mentions + global concept clusters) | hybrid | Yes | JSON (4 sections) |
| MiniRAG | Flat graph | light | No | CSV (3 sections) |
| PathRAG | Two-tier hierarchy (high-level + low-level) | hybrid | No | CSV (5 sections) |
| HyperGraphRAG | Hyperedge graph | hybrid | No | CSV (3 sections) |
Choosing a Backend
Different clinical and infrastructure requirements call for different backends. Use the guidance below to make an informed choice.- Use LightRAG when you need the richest knowledge graph traversal and already have (or can run) a Neo4j instance. Its dual-level graph and hybrid retrieval mode give the most comprehensive entity- and concept-level coverage.
- Use MiniRAG in resource-constrained settings — such as low-memory environments or rapid prototyping — where you need a working Graph RAG pipeline without Neo4j and with minimal LLM-call overhead.
- Use PathRAG when chain-of-evidence reasoning is a priority. Path-based retrieval is particularly well suited to clinical workflows that require following a chain from symptom to mechanism to treatment to contraindication.
- Use HyperGraphRAG when your clinical reasoning involves syndromes or multi-factorial conditions where a single relationship implicates several entities simultaneously — for example, a syndrome that jointly involves multiple symptoms, biomarkers, and treatments.
Backend Pages
LightRAG
Dual-level knowledge graph with hybrid retrieval. The richest backend — and the only one that requires Neo4j.
MiniRAG
Lightweight flat-graph backend with light retrieval mode, optimised for constrained environments.
PathRAG
Two-tier hierarchical graph with path traversal — ideal for chain-of-evidence clinical reasoning.
HyperGraphRAG
Hyperedge-aware graph that captures many-to-many group relationships among medical concepts.