A collection of AI/ML engineering projects built for production.

๐Ÿ›ก๏ธ
Privacy AI
Apr 2026

GuardPass

A stateful privacy middleware that intercepts user prompts, detects and anonymizes PII before forwarding to an LLM, then restores original values in the response.

  • โ€“Token-mapped PII restoration across multi-turn conversations
  • โ€“Pluggable GPT-4o / GPT-4o-mini backends with clean schemas
  • โ€“Full-stack: Python backend + React frontend + audit dashboard
PythonReactGPT-4oPII DetectionREST API
View on GitHub
โ˜๏ธ
Cloud ยท AWS
May 2026

AI-Powered Secure Web App

Production-ready, security-conscious web application on AWS combining AI capabilities with cloud-native architecture for scalability and resilience.

  • โ€“Deployed on AWS with fault-tolerant, scalable infrastructure
  • โ€“Secure authentication and session management
  • โ€“End-to-end from frontend to cloud deployment
AWSCloud ArchitectureSecurityAI
View on GitHub
๐ŸŽฌ
Multi-Modal AI ยท Published
Research Published
Jun 2024

Video Summarization Storyboard

End-to-end agentic pipeline orchestrating Gemini 1.5, DALL-E 3, and Midjourney as callable tools to generate visual storyboards from YouTube videos.

  • โ€“Published in IJCRT Vol. 12, Issue 6, June 2024
  • โ€“Orchestrated 3 AI APIs with structured schemas
  • โ€“Python ingestion for large-scale unstructured video datasets
PythonGemini 1.5DALL-E 3Prompt Engineering
View on GitHub
๐Ÿ‘๏ธ
Deep Learning ยท 94% Accuracy
Mar 2024

Violence Detection in Videos

CNN-based classification model for real-time violence detection in video streams with systematic architecture tuning on large-scale datasets.

  • โ€“94% accuracy via systematic CNN architecture tuning
  • โ€“Full pipeline: ingestion โ†’ preprocessing โ†’ feature extraction โ†’ eval
  • โ€“Reproducible, documented workflows for scalable automation
TensorFlowPyTorchCNNOpenCVPython
๐Ÿš†
Real-Time ML ยท 96% Precision
2024

Railway Trespasser Detection

Real-time ML model for automated trespasser detection using terrain video streams, integrated into a production pipeline for automated action-taking.

  • โ€“96% precision on live terrain video streams
  • โ€“Production pipeline with automated alert system
  • โ€“End-to-end ML in safety-critical visual environments
PythonDeep LearningComputer VisionReal-Time ML