kimsb2429

                 ingestion  ────────────────────  query  ──────────────────
                 ┌─────┐                          ┌─────┐
                 │     │                          │     │
   Enterprise:  git── ▼     ▼ ── pgvector ──▶ search ──▶ ▼ ── chunks back
   (work)       clone        cloud-embed         (vector/text/      (reconstruct
                              model               hybrid)            from overlap)


   Demo:        crawl ▼     ▼ ── pgvector ──▶ MCP ──▶  ▼ ── rerank ──▶ generate
   (this)       +     local mxbai            tool      query                │

kimsb2429

                     ┌──────────────────────────────┐
                     │  API Gateway / orchestrator  │
                     │   (its own repo, FastAPI)    │
                     └──────────────────────────────┘
                        │       │       │       │
          ┌─────────────┘       │       │       └───────────┐
          ▼                     ▼       ▼                   ▼
   ┌──────────────┐       ┌─────────────────┐      ┌──────────────────┐
   │ search-svc   │       │ rerank-svc      │      │ generation-svc   │
   │ (separate    │       │ (separate repo) │      │ (separate repo)  │

kimsb2429

INGESTION (one-time per corpus):

   crawler           preprocessor     structure-aware     mxbai-embed-large
   (HTML + PDF) ──▶  (source-     ──▶ chunker         ──▶ (local, 1024-dim)
                     specific)         (preserves
                                        table colspan
                                        /rowspan)
                                                                  │
                                                                  ▼
                                                          ┌───────────────┐