Checking…

System

Architecture

Production-minded RAG for knowledge and support — pipeline exposed, not hidden.

                   ┌───────────────┐
                   │  Documents    │
                   └───────┬───────┘
                           ↓
                   ┌───────────────┐
                   │    Parse      │
                   └───────┬───────┘
                           ↓
                   ┌───────────────┐
                   │    Chunk      │
                   └───────┬───────┘
                           ↓
                   ┌───────────────┐
                   │  Embeddings   │
                   └───────┬───────┘
                           ↓
        ┌─────────────────────────────────┐
        │       Hybrid Retrieval          │
        │  Dense ─┐                         │
        │         ├─ RRF → Rerank           │
        │  BM25 ──┘                         │
        └────────────────┬────────────────┘
                         ↓
                  ┌──────────────┐
                  │   Generate   │
                  └──────┬───────┘
                         ↓
               ┌───────────────────┐
               │  Ground / Refuse    │
               └─────────┬───────────┘
                         ↓
                 Answer + Citations

Pipeline stages

  1. 01

    Ingest

    Fetch docs from catalogued public sources.

  2. 02

    Parse

    Structure-preserving HTML, MDX, and Markdown parsers.

  3. 03

    Chunk

    Structural chunks tuned for technical documentation.

  4. 04

    Embed

    BGE-small dense vectors with optional pgvector store.

  5. 05

    Understand

    Route, rewrite, optional HyDE for dense query.

  6. 06

    Retrieve

    Dense cosine + BM25 → RRF → feature rerank.

  7. 07

    Generate

    Extractive, Ollama, or Bedrock — all grounded.

  8. 08

    Ground

    Cite sources or refuse when evidence is insufficient.

Retrieval

Dense + BM25 hybrid with reciprocal rank fusion and lexical reranking.

Generation

Extractive by default (free). Ollama or Bedrock optional behind the same interface.

Storage

In-memory embedding cache at deploy. Optional Postgres + pgvector for scale.

Deployment

Vercel (UI) + Railway (API). Docker Compose for local full stack.