AI-Driven Integrated Prison Management System
Work Summary
A microservices-based prison management platform for Sri Lanka with four core components: Inmate Records, Violation Tracking, AI Rehabilitation and Inmate Wellness Monitoring.
The Challenge
Prison administrations rely heavily on paper records and manual assessments, which makes it hard to track inmate behaviour, rehabilitation progress and mental wellbeing. This system brings these processes together and adds AI-driven insights on top.
- AI Rehabilitation component built with XGBoost and Scikit-learn.
- Facial emotion and gender detection from live WebRTC video using a CNN (YOLOv11) trained on custom datasets.
- Voice emotion analysis on microphone audio streams using quantized Word2Vec (CBOW) to detect vocal distress.
- NLP sentiment analysis of psychological questionnaires using HuggingFace Transformers (DistilBERT/BART).
- Medical RAG system that digitizes handwriting with GLM-OCR and embeds historical PDFs into ChromaDB for semantic search.
The Solution
Splitting the platform into independent services let each AI component evolve on its own, while giving prison staff a single place to manage records, violations, rehabilitation and wellness.
Technologies
PythonMicroservicesYOLOv11XGBoostScikit-learnHuggingFace TransformersWebRTCChromaDBRAG
