README.md

metadata

sdk: gradio
title: Edge Node Diagnostics Agent
emoji: 🔬
colorFrom: blue
colorTo: purple
sdk_version: 6.14.0
python_version: '3.13'
app_file: app.py
pinned: false
tags:
  - ml-intern
  - diagnostics
  - edge-computing
  - face-recognition

🔬 Edge Node Diagnostics Agent

A specialized AI assistant for troubleshooting and optimizing the SmartClass face recognition pipeline running on classroom edge devices (Raspberry Pi 5 / Intel NUC).

Features

💬 Conversational Diagnostics

Ask natural language questions about your edge node's health and get detailed, actionable diagnosis with specific configuration changes.

🔧 Diagnostic Tools

The agent automatically queries relevant subsystems:

• Pipeline Status — Overall health and component status
• Prometheus Metrics — Real-time performance data
• Detection Analysis — SCRFD model performance
• Recognition Analysis — MobileFaceNet + FAISS status
• Tracking Analysis — ByteTrack subsystem
• Event Transmission — Redis/HTTP/SQLite delivery
• Identity Engine — Evidence accumulation status
• Error Logs — Recent warnings and errors
• Configuration — Current YAML settings

📊 Live Metrics Dashboard

Real-time (simulated) Prometheus metrics visualization updating every 5 seconds.

⚙️ Configuration Panel

Interactive sliders for tuning all pipeline parameters with immediate effect.

💥 Fault Injection

Simulate common failure scenarios to test diagnostic capabilities:

• Low FPS (heavy processing load)
• Redis Disconnect (network failure)
• No Recognitions (threshold misconfiguration)
• High Temperature (thermal throttling)
• Tracking Chaos (ghost tracks)

Pipeline Architecture

Camera → SCRFD Detection → ByteTrack → Quality Gate → Face Align → MobileFaceNet → FAISS Search → Identity Evidence Engine → Redis/HTTP/SQLite

Example Questions

• "The edge node is only getting 2 FPS, how do I optimize?"
• "Faces are detected but never recognized"
• "Tracking IDs keep switching between the same person"
• "The offline queue keeps growing, events aren't being sent"
• "How do I tune the identity evidence engine?"

Technical Stack

• Detection: SCRFD (ONNX Runtime)
• Tracking: ByteTrack
• Recognition: MobileFaceNet / AdaFace ensemble
• Search: FAISS (cosine similarity)
• Transmission: Redis Streams → HTTP fallback → SQLite offline queue
• Monitoring: Prometheus metrics on port 9100