██████╗ ██████╗  ██████╗ ███╗   ██╗███████╗██╗  ██╗
 ██╔════╝██╔═══██╗██╔════╝ ████╗  ██║██╔════╝╚██╗██╔╝
 ██║     ██║   ██║██║  ███╗██╔██╗ ██║█████╗   ╚███╔╝
 ██║     ██║   ██║██║   ██║██║╚██╗██║██╔══╝   ██╔██╗
 ╚██████╗╚██████╔╝╚██████╔╝██║ ╚████║███████╗██╔╝ ██╗
  ╚═════╝ ╚═════╝  ╚═════╝ ╚═╝  ╚═══╝╚══════╝╚═╝  ╚═╝

Knowledge, unchained from the cloud.

COGNEX is a fully offline factual answering engine that delivers precise, structured answers to natural language queries — without any internet connection or external APIs.

---

Screenshots

Hero Landing

Search Results — Intelligence View

Dashboard

Browse by Category

No Results State

Archives View

Telemetry View

Nodes View

System View

---

Features

• Offline-First — Zero API calls, zero internet required. Everything runs locally.
• BM25 Retrieval — Industry-standard ranking algorithm with field-weighted scoring
• Intent Classification — 7 intent types (Definition, Formula, Date, Person, Place, Comparison, Factual)
• 461 Curated Facts — Across 8 categories: Science, Mathematics, History, Geography, Technology, Language, Health, General
• Sub-millisecond Retrieval — Pre-built inverted index with alias fast-path
• Command Center UI — Stitch-designed dashboard with 5 views
• Confidence Scoring — Visual confidence arc with color-coded reliability indicators
• Docker Ready — Multi-stage build, runs anywhere

Architecture

graph TB
    subgraph Frontend["Frontend — Stitch-Designed UI"]
        HERO["Hero Landing Page"]
        DASH["Command Center Dashboard"]
        HERO -->|"Search Query"| DASH
        DASH --- INT["Intelligence"]
        DASH --- ARC["Archives"]
        DASH --- TEL["Telemetry"]
        DASH --- NOD["Nodes"]
        DASH --- SYS["System"]
    end

    subgraph API["FastAPI Server"]
        Q["/api/query"]
        S["/api/stats"]
        C["/api/categories"]
        H["/api/health"]
    end

    subgraph Engine["Retrieval Engine"]
        IA["Intent Analyzer<br/>7 Intent Types"]
        RT["BM25 Retriever<br/>Field-Weighted Scoring"]
        IX["Inverted Index<br/>4107 Tokens"]
        IA -->|"QueryAnalysis"| RT
        RT -->|"Token Lookup"| IX
    end

    subgraph Data["Knowledge Base"]
        KB["knowledge_base.jsonl<br/>461 Entries · 8 Categories"]
    end

    Frontend -->|"HTTP Requests"| API
    Q --> IA
    S --> IX
    C --> KB
    IX -->|"Pre-built Index"| KB

    style Frontend fill:#0d0d1f,stroke:#00e5ff,color:#f0f2f5
    style API fill:#0d0d1f,stroke:#7c4dff,color:#f0f2f5
    style Engine fill:#0d0d1f,stroke:#00e676,color:#f0f2f5
    style Data fill:#0d0d1f,stroke:#ffab00,color:#f0f2f5

Loading

Query Flow

sequenceDiagram
    participant U as User
    participant F as Frontend
    participant A as FastAPI
    participant I as Intent Analyzer
    participant R as BM25 Retriever
    participant K as Knowledge Base

    U->>F: Types query
    F->>A: POST /api/query
    A->>I: analyze(query)
    I-->>A: QueryAnalysis (intent, entity, tokens)
    A->>R: search(analysis, top_k=5)
    R->>R: Check alias fast-path
    R->>K: BM25 token lookup
    K-->>R: Matching documents + scores
    R->>R: Apply intent boost + category boost
    R-->>A: Top 5 SearchResults
    A-->>F: JSON response (results, confidence, timing)
    F-->>U: Render result card + animations

Loading

Quick Start

Local

pip install fastapi uvicorn[standard]
python -c "from indexer import CognexIndexer; CognexIndexer().build_and_save('knowledge_base.jsonl', 'cognex_index.pkl')"
uvicorn main:app --host 0.0.0.0 --port 8000

Open http://localhost:8000

Docker

docker build -t cognex .
docker run -p 8000:8000 cognex

Dashboard Views

View	Description
Intelligence	Main search interface with confidence scoring and related results
Archives	Browse all 461 entries organized by category
Telemetry	Real-time query performance metrics and analytics
Nodes	Knowledge graph visualization of entry connections
System	Health monitoring, index info, and system configuration

Tech Stack

Component	Technology
Backend	Python 3.12, FastAPI, Uvicorn
Retrieval	Custom BM25 engine (stdlib only)
NLP	Rule-based intent classifier with suffix stemmer
Knowledge Base	JSONL (461 curated entries)
Frontend	Vanilla HTML/CSS/JS, Google Stitch design
Container	Docker (multi-stage build)

API Endpoints

Method	Endpoint	Description
POST	/api/query	Submit a factual query
GET	/api/stats	Knowledge base statistics
GET	/api/categories	Category list with counts
GET	/api/health	System health check

Performance

Metric	Value
Average Query Time	<1ms
Intent Accuracy	100% (55/55 tests)
Top-3 Hit Rate	100%
Index Build Time	<1s
Knowledge Base	461 entries, 4107 tokens

Project Structure

cognex/
├── main.py              # FastAPI server + API routes
├── indexer.py           # BM25 inverted index builder
├── retriever.py         # Search engine with alias fast-path
├── intent.py            # NLP intent classifier + stemmer
├── knowledge_base.jsonl # 461 curated factual entries
├── test_cognex.py       # 55 validation tests
├── static/
│   ├── index.html       # Stitch-designed command center UI
│   └── favicon.svg      # COGNEX hexagonal logo
├── screenshots/         # Application screenshots
├── Dockerfile           # Multi-stage Docker build
├── docker-compose.yml   # One-command deployment
├── requirements.txt     # fastapi + uvicorn only
├── glitch.json          # Glitch deployment config
└── README.md

License

MIT License — see LICENSE for details.