SensorLens — Manufacturing Anomaly Detection & LLM Explanation

AI-powered system that detects anomalous sensor readings in manufacturing equipment and explains them in plain English using LLaMA 3 70B.

Live Demo: [link — add after deploy]

Architecture

┌─────────────────────────────┐
│   Streamlit Cloud (Frontend)│
│   frontend/app.py           │
└──────────────┬──────────────┘
               │ HTTP (REST)
┌──────────────▼──────────────┐
│ Oracle Always Free (Backend)│
│   ├── detector.py           │
│   │    └── Isolation Forest │
│   └── explainer.py          │
│        └── Groq API         │
│             (LLaMA 3 70B)   │
└─────────────────────────────┘

Tech Stack

Component	Technology
Backend framework	Python 3.11+, FastAPI
Anomaly detection	scikit-learn — Isolation Forest
LLM inference	Groq API — LLaMA 3.3 70B Versatile
Frontend	Streamlit
Dataset	AI4I 2020 Predictive Maintenance (10k rows)
Backend deploy	Oracle Always Free
Frontend deploy	Streamlit Cloud

Features

• Configurable Isolation Forest with full parameter control
• Auto-explains all anomalies via LLaMA 3 70B in batches
• Natural language querying over anomaly data
• 4 visualization types: scatter, distributions, heatmap, time series
• Ground truth comparison against 5 labelled failure types
• CSV export of anomaly results

Local Setup

git clone https://github.com/411sst/sensorlens
cd sensorlens

# Create and activate a virtual environment
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

# Install dependencies
pip install -r backend/requirements.txt
pip install -r frontend/requirements.txt

# Configure environment
cp .env.example .env
# Edit .env — paste your GROQ_API_KEY (free key at https://console.groq.com)
# Leave BACKEND_URL=http://localhost:8000 for local dev

# Download the dataset from Kaggle and place it at data/ai4i2020.csv
# https://www.kaggle.com/datasets/stephanmatzka/predictive-maintenance-dataset-ai4i-2020

# Start the backend (must run from inside backend/ due to relative imports)
cd backend
uvicorn main:app --reload

# In a new terminal — run from the project root
streamlit run frontend/app.py

Oracle Deployment Notes

• Deploy backend on an Oracle Always Free Ubuntu VM.
• Expose the FastAPI app through a reverse proxy (for example, Caddy or Nginx).
• Set GROQ_API_KEY on the VM environment.
• Point Streamlit Cloud to your backend URL in secrets:

BACKEND_URL = "https://api.yourdomain.com"

• Open inbound ports in Oracle networking: 22, 80, 443.
• Verify backend health after deploy:

curl https://api.yourdomain.com/health

Dataset

AI4I 2020 Predictive Maintenance Dataset — https://www.kaggle.com/datasets/stephanmatzka/predictive-maintenance-dataset-ai4i-2020

10,000 rows of manufacturing sensor readings with 5 failure type labels.