SentraGrid is an advanced, AI-powered industrial safety platform designed to prevent catastrophic plant accidents by identifying compound risk patterns that traditional, single-sensor alarms completely miss.
By unifying real-time gas/temperature telemetry, geospatial worker positions, and active work permits into a centralized intelligence system, SentraGrid eliminates the critical visibility gaps responsible for 73% of plant near-misses (according to the FICCI Industrial Safety Report).
SentraGrid connects remote physical telemetry and operational operational metadata to a real-time risk evaluation engine, streaming notifications and AI safety insights straight to the control room.
graph TD
%% Telemetry Sources
SubGraph1[IoT Telemetry & Permits]
Sensors["Sensors (H₂S, CO, Temp)"] --> SubGraph1
Permits["Work Permits (Hot Work)"] --> SubGraph1
Workers["Worker GPS Positions"] --> SubGraph1
%% Backend Services
SubGraph2[FastAPI Core Backend]
SubGraph1 --> |REST / POST| RiskEngine["Compound Risk Engine"]
RiskEngine --> |Retrieval & Query| RAGService["pgvector RAG Service"]
%% AI Model Services
Groq["Groq API (Llama 3.3)"] <--> |Synthesis Q&A| RAGService
HFModel["all-MiniLM-L6-v2"] <--> |Embeddings| RAGService
%% Database
Supabase["Supabase DB (PostgreSQL)"] <--> |pgvector Match RPC| RAGService
RiskEngine --> |Store Readings/Alerts| Supabase
%% Real-time Sync
Supabase --> |Realtime WAL PubSub| SSE["Server-Sent Events (SSE)"]
%% Frontend Console
SubGraph3[Next.js Cyberpunk Dashboard]
SSE --> |Live Streams| Hook["useRealtimeStream() Hook"]
Hook --> Map["Dynamic Plant Heatmap (SVG)"]
Hook --> Alerts["Predictive Risk Insight Cards"]
Hook --> Compliance["Compliance & Safety Log"]
Traditional alarms only trigger when a single sensor breaches a critical threshold (e.g., CO > 25 ppm). SentraGrid solves the "sub-threshold hazard" problem by correlating multiple telemetry streams:
- Combines H₂S (7.5 ppm) + CO (24 ppm) + Active Hot Work Permit in the same zone.
- Calculates a compound risk score and preempts accidents with an early warning time window.
- Calculates and displays Predictive Lead-Time Estimates (e.g., "Incident imminent in ~35 mins") and Confidence Scores.
Control room operators can test safety scenarios in real-time. Dragging temperature or toxic gas sliders dynamically updates the zone's risk color on the plant map and triggers immediate compound alerts.
- Submits custom sensor readings via the
POST /api/sensors/readingREST API. - Instantly re-calculates safety scores on the backend.
Supports instant geospatial interface switching between two profiles:
- Visakhapatnam Steel Complex: Renders Coke-Oven Batteries (1-4), Gas Holder Station, and By-product Recovery Units.
- Jamnagar Petrochemical Refinery: Renders Crude Distillation (CDU), Fluid Catalytic Cracking (FCCU), Hydrogen Generation (HGU), Refinery Tank Farm, and Central Control Station SVG nodes and connection matrices.
Allows plant managers to query historical logs and safety standards (OISD, Factory Act) using semantic similarity vector search:
sequenceDiagram
autonumber
actor Operator as Safety Officer
participant FE as Next.js UI
participant BE as FastAPI Server
participant Embed as SentenceTransformer
participant DB as Supabase pgvector
participant LLM as Groq (Llama 3.3)
Operator->>FE: Ask: "Have we had gas incidents in confined spaces before?"
FE->>BE: GET /api/rag/query?question=...
BE->>Embed: encode(question)
Embed-->>BE: 384-dim Vector
BE->>DB: RPC: match_incident_reports(vector, top_k=5)
DB-->>BE: Relevant incident chunks & similarity scores
BE->>LLM: call_llm(context_chunks + question)
LLM-->>BE: Synthesized, cited markdown answer
BE-->>FE: Return JSON {answer, sources}
FE->>Operator: Display answer + source links
A transparent compliance ledger showing all active permit overrides, safety justifications, timestamps, and safety officer approvals to prevent undocumented hazardous entries.
SentraGrid uses standard HTML5 Server-Sent Events (SSE) to broadcast live plant telemetry changes to the frontend client.
- Stream Endpoint:
GET /api/simulator/stream - Event Channels Broadcasted:
alerts: Open alerts generated by the compound risk engine.sensor_readings: Real-time gas concentrations, temperatures, and pressures.worker_positions: Real-time coordinates of workers within facility zones.permits: Live authorization and status changes of hot-work or confined space permits.
Allows operators to inject mock readings into any active sensor to perform safety stress-testing.
- Route:
POST /api/sensors/reading - Payload Format:
{ "sensor_id": "sensor-h2s-zone-1", "value": 8.5 } - Behavior: Writing a custom value automatically triggers a backend evaluation sweep. If the compound risk index breaches limits, a SSE message is instantly pushed to the dashboard to update zone states and UI map gradients.
The main dashboard includes an ROI analytics card tracking the economic and operational value generated by SentraGrid:
- Checklists Saved: Number of manual paper checklists automated by digital integrations.
- Early Warning Lead-Time: Average warning margin provided before a critical breach (typically +35 minutes early warning advantage).
- Unified Safety Databases: Combines telemetry databases, permit-to-work servers, and geolocation databases into one view, reducing handoff latency by 100%.
In 2022, concurrent hot-work permits and leaking seals during coke-oven door replacement caused a toxic gas leak. Standard alarms failed to sound because H₂S was at 7.5 ppm (limit 10) and CO was at 24 ppm (limit 25).
- Without SentraGrid: Isolated databases and sub-threshold metrics resulted in a critical exposure incident.
- With SentraGrid: The compound risk engine detects the concurrent hazards, predicts the leak 35 minutes in advance, and raises a critical alarm.
| Layer | Technology | Key Usage |
|---|---|---|
| Frontend | React / Next.js (TypeScript) | Cyberpunk UI Dashboard, SSE Stream, Dynamic SVGs |
| Backend | FastAPI (Python) | High-performance async routers, SIMOPS simulator |
| Database | Supabase (Postgres) | pgvector storage, match RPC, Realtime SSE |
| Embedding Model | SentenceTransformer (all-MiniLM-L6-v2) |
Local 384-dimension semantic vector generation |
| LLM Orchestration | Groq API (llama-3.3-70b-versatile) |
Contextual reasoning, predictive analysis, RAG Q&A |
To run SentraGrid in production mode (MOCK_MODE=false), configure these variables in /backend/.env:
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_SERVICE_KEY=your-secret-service-role-key
GROQ_API_KEY=gsk_your-groq-api-key
GROQ_MODEL=llama-3.3-70b-versatile
EMBEDDING_MODEL=all-MiniLM-L6-v2
MOCK_MODE=false
CORS_ORIGINS=http://localhost:3000- Enable the
vectorextension in your Supabase project. - Run the SQL schema script in 001_initial_schema.sql in your Supabase SQL Editor to initialize tables and the vector similarity search RPC.
cd backend
python -m venv venv
.\venv\Scripts\activate
pip install -r requirements.txt
python app/seed_db.py # Seeds the Supabase database with embeddings
python -m uvicorn app.main:app --reload --port 8000cd frontend
npm install
npm run devOpen http://localhost:3000 in your browser to view the SentraGrid Operations Center.