Agentry — Automated Production Agents
A production-grade Python AI agent framework designed for building, deploying, and managing autonomous AI agents. Features built-in CI/CD pipelines, MCP (Model Context Protocol) integration, and support for both TUI and GUI interfaces.
Key Features
- Custom AI agent framework with modular architecture
- Built-in CI/CD pipelines for seamless deployment
- MCP integration for enhanced model communication
- Support for Terminal UI (TUI) and Graphical UI (GUI)
- Production-ready with logging, monitoring, and error handling
MARG — Traffic Signal Waiting Time Reduction
A real-time traffic management system developed for Smart India Hackathon 2024. Uses YOLO object detection and SAHI (Slicing Aided Hyper Inference) for accurate vehicle detection and dynamic signal scheduling to optimize traffic flow.
Key Features
- Real-time vehicle detection using YOLOv8
- SAHI integration for improved small object detection
- Dynamic signal timing based on traffic density
- Dashboard for traffic monitoring and analytics
- Grand Finalist at Smart India Hackathon 2024
Advanced Image Captioning
A high-performance image captioning system using ResNet-101 as the encoder and Bahdanau Attention mechanism for the decoder. Achieves state-of-the-art BLEU scores with BLEU-1: 0.61.
Key Features
- ResNet-101 encoder for robust feature extraction
- Bahdanau Attention mechanism for focused captioning
- BLEU-1 score of 0.61 on benchmark datasets
- Beam search decoding for improved caption quality
- Visualization of attention weights over images
GPT-2 from Scratch
An educational project involving the complete implementation and pre-training of a GPT-2 style transformer-based language model using PyTorch. Built from first principles to deeply understand the transformer architecture.
Key Features
- Complete transformer architecture implementation
- Multi-head self-attention mechanism
- Layer normalization and positional encoding
- Custom tokenizer and data preprocessing
- Training pipeline with gradient accumulation