Industrial Knowledge Intelligence Engine - Turning fragmented engineering evidence into a connected operating memory using Graph RAG & Generative AI.
Built for the ET Gen AI 2.0 Hackathon
Problem Statement 8. AI for Industrial Knowledge Intelligence: Unified Asset & Operations Brain
In complex industrial facilities, engineers and maintenance technicians spend up to 35% of their working hours manually searching across fragmented siloes — P&IDs, equipment manuals, incident logs, work orders, and ISO regulatory standards.
OnBrain bridges this gap by unifying unstructured engineering documents into a Graph RAG (Retrieval-Augmented Generation) knowledge system. By fusing Neo4j Knowledge Graphs (for equipment relationship mapping) with ChromaDB Vector Stores (for semantic chunk search) and Google Gemini (for reasoning & RCA synthesis), OnBrain delivers instant, verifiable, source-cited answers to frontline technicians.
| ❌ The Industrial Problem | ✅ The OnBrain Solution |
|---|---|
| Fragmented Siloes: Critical history is trapped across separate databases and PDF archives. | Unified Operating Memory: Ingests PDFs, CSVs, JSONs, and TXT files into a single graph + vector index. |
| LLM Hallucinations: Standard RAG often invents non-existent equipment specs or maintenance steps. | Graph-Backed Grounding: Every AI response requires explicit document citations and confidence scoring. |
| Lost Expertise: Decades of tribal knowledge vanish when senior engineers retire. | Self-Learning Knowledge Graph: Automatically links equipment tags (e.g., P-204) with historical failure events. |
| Downtime Losses: Slow root-cause analysis causes millions in unplanned operational outages. | Instant Root Cause Analysis: Automated RCA agent synthesizes failure history, OEM manuals, and inspection logs. |
- Conversational Intelligence: Ask complex queries like "What compliance gaps exist in my documents?" or "Show failure history for pump P-204".
- Source Citations & Confidence: Every response lists exact source documents, page numbers, and confidence metrics (
High,Medium,Pending). - Clean Responsive UX: Optimized for field mobile viewports as well as widescreen engineering control panels.
- Multi-Taxonomy Intake: Auto-detects and indexes manuals, work orders, inspection reports, P&ID drawings, and regulatory standards.
- Entity Extraction: Automatically extracts equipment tags, operational parameters, dates, and personnel.
- Triple Database Sync: Stores metadata in Firestore, vector embeddings in ChromaDB, and entity lineages in Neo4j Aura.
- High-Density Corpus Table: Full visibility into indexed files, chunk counts, extracted entities, and synchronization status.
- Taxonomy Filtering: One-click filtering by document type (Manual, Work Order, Inspection Report, Regulation, P&ID).
- Semantic Coverage Analysis: Search risk scenarios or procedure requirements against indexed regulatory manuals.
- Quick-Scan Prompts: One-click assessment chips for pressure vessel inspection intervals, pump vibration limits, LOTO safety, and ISO 9001 compliance.
graph TD
%% Styling
classDef frontend fill:#2d3748,stroke:#4fd1c5,stroke-width:2px,color:#fff
classDef backend fill:#2c5282,stroke:#63b3ed,stroke-width:2px,color:#fff
classDef cloud fill:#744210,stroke:#f6e05e,stroke-width:2px,color:#fff
classDef ai fill:#553c9a,stroke:#b794f4,stroke-width:2px,color:#fff
%% Nodes
UI[<b>Frontline Technician / UI</b><br/>React 18 + Vite + Tailwind]:::frontend
API[<b>FastAPI Backend</b><br/>Python 3.11 + Pydantic]:::backend
subgraph Databases [Triple-Database Architecture]
NEO[<b>Knowledge Graph</b><br/>Neo4j Aura Cloud<br/><i>Equipment Lineage & Events</i>]:::cloud
FS[<b>Metadata Store</b><br/>Firestore Cloud<br/><i>Auth & Document State</i>]:::cloud
VDB[<b>Vector Store</b><br/>ChromaDB Local<br/><i>Semantic Chunks & Retrieval</i>]:::cloud
end
LLM[<b>Reasoning & Agent Engine</b><br/>Google Gemini / Groq APIs<br/><i>RCA & Source Citation</i>]:::ai
%% Relationships
UI -- "REST / Streaming API" --> API
API -- "Graph Traversal" --> NEO
API -- "Doc State & Auth" --> FS
API -- "Vector Search" --> VDB
API -- "RAG Prompts" --> LLM
LLM -. "Reasoning context" .-> Databases
The architecture is fully migrated to managed cloud services for a lightweight local setup. Docker is no longer required for running the database services.
- Node.js v18+ and npm v9+
- Python 3.11+
- Firebase Project (with Firestore and Authentication enabled)
- Neo4j Aura (Free Tier graph database instance)
git clone https://github.com/Tetra4ge/OnBrain.git
cd OnBrainCreate a .env file in the backend/ directory with your cloud credentials:
ENVIRONMENT=development
PORT=8000
# Cloud Graph DB
NEO4J_URI=neo4j+s://<your-aura-id>.databases.neo4j.io
NEO4J_USER=neo4j
NEO4J_PASSWORD=your_secure_password
# Local Vector DB
CHROMA_DATABASE=Onbrain
# LLMs
GROQ_API_KEY=your_groq_api_key
GEMINI_API_KEY=your_gemini_api_key
# Cloud Metadata
FIREBASE_PROJECT_ID=your_firebase_project_id
FIREBASE_SERVICE_ACCOUNT_PATH=./service-account.jsonCreate a .env file in the frontend/ directory:
VITE_MODE=development
VITE_API_DEV_URL=http://localhost:8000
VITE_FIREBASE_API_KEY=your_firebase_api_key
VITE_FIREBASE_AUTH_DOMAIN=your_project.firebaseapp.com
VITE_FIREBASE_PROJECT_ID=your_projectcd backend
python -m venv venv
# On Windows:
.\venv\Scripts\activate
# On Linux/macOS:
source venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000API interactive docs available at: http://localhost:8000/docs
cd ../frontend
npm install
npm run devWeb dashboard available at: http://localhost:5173
- Zero-Hallucination Grounding: Unlike pure LLM chatbots, OnBrain cross-checks every generated response against both vector similarity search (ChromaDB) and graph lineage (Neo4j), providing source citations for complete auditability.
- True Industrial Value: Tackles a multi-billion dollar operational problem (unplanned industrial downtime and lost engineering context).
- Production-Ready UX: Responsive mobile & desktop interface designed with dark industrial aesthetics, responsive mobile navigation, and non-scrollable focus viewports.
- Hybrid Graph RAG Architecture: Combines semantic embeddings with graph entity relationships for deep multi-hop reasoning.
Built Team TetraFourge
© 2026 OnBrain. All rights reserved.