---

About

AI/ML Engineer passionate about building end-to-end AI systems. I believe in understanding AI by implementing from scratch — from Vision Transformers to multi-agent architectures.

Focus: Computer Vision • Research Implementations • GenAI & Agents • MLOps & LLMOps Cloud Deployment

---

Tech Stack

Programming Languages

ML/AI Frameworks

Computer Vision

Supervision YOLOv5-v8 Object Detection Image Segmentation DeepSORT ByteTrack Optical Flow

GenAI & LLM

LangGraph AG2 (AutoGen) Agno RAG DeepSeek Prompt Engineering OpenAI API

Data Science

Seaborn Scikit-learn Statistical Analysis Feature Engineering A/B Testing

Development Tools

Streamlit FastAPI Flask Gradio

Cloud & Deployment

EC2 S3 Lambda CI/CD MLflow DVC

Databases

Pinecone Chroma FAISS Vector Databases

---

Featured Projects

Tennis Vision ★ 23 Real-time tennis match analysis system with advanced computer vision YOLOv8 custom-trained on 1000+ annotated frames ByteTrack for robust multi-object tracking CNN-based court homography detection Shot classification: 85%+ accuracy Player detection: 95% | Ball tracking: 88% Mini-court tactical visualization Real-time speed & distance analytics Python YOLOv8 OpenCV PyTorch ByteTrack Supervision	Field Fusion Multi-sport CV analysis pipeline for football/soccer YOLOv8 + DeepSORT tracking pipeline K-means team identification using HSV color Optical flow camera compensation Tactical heatmap generation Player possession & movement analysis Perspective transformation for top-view Ball control & pass detection Python YOLO DeepSORT OpenCV Supervision
Histopathology Analysis Deep learning medical imaging system for cancer detection Binary classification: benign/malignant Grad-CAM heatmaps for model explainability OpenCV feature extraction pipeline Auto-generated PDF diagnostic reports Transfer learning with pretrained CNNs Flask web interface for predictions TensorFlow OpenCV Flask Grad-CAM Keras	MolecuQuest AI-powered molecular research & drug discovery platform NVIDIA MolMIM for molecule generation RDKit 3D molecular visualization PubChem API integration for data CMA-ES optimization + QED scoring Interactive molecular property explorer Firebase backend for user sessions TypeScript Next.js Firebase RDKit NVIDIA NIMs
QuantaAI Conversational AI with real-time web search capabilities LangGraph agentic architecture Real-time web search integration GPT-4 powered intelligent responses Live search visualization UI Multi-turn conversation memory FastAPI backend + Next.js frontend LangGraph Next.js FastAPI GPT-4 OpenAI API	DeepGuard Production-ready Deepfake Detection MLOps Pipeline EfficientNet-based image classifier DVC pipeline for data versioning MLflow experiment tracking on DagsHub Automated CI/CD with GitHub Actions Docker containerized deployment Deployed on Hugging Face Spaces TensorFlow DVC MLflow Docker CI/CD AWS S3 & EC3 EKS Deployment Prometheus Grafana

---

Research Implementations

Building architectures from scratch to truly understand them.

Paper	Implementation	Key Details
Vision Transformer	ViT	16×16 patch embedding, 12-layer encoder, multi-head attention
LoRA & QLoRA	PyTorch-LoRA	Low-rank adaptation, 4-bit quantization, 83% VRAM reduction
Vision-Language Models	VLMverse	PaLiGemma, SigLIP, cross-modal attention fusion

---

Multi-Agent Systems

Project	Description
AgentForge ★ 2	Multi-agent orchestration with CrewAI, LangGraph, PhiData
Travel Planner	4-agent system: Flight, Hotel, Itinerary, Budget
Google ADK	Production agent patterns with Google ADK

---

Learning Repos

Comprehensive implementations from fundamentals to SOTA:

• ML & Deep Learning — Classical ML, CNNs, RNNs, Transformers, NLP
• Computer Vision — RCNN, YOLO, U-Net, Mask R-CNN, tracking
• PyTorch Deep Dive — Autograd, custom layers, DDP, AMP
• RAG Systems — Vector DBs, chunking, retrieval, production patterns

---

GitHub Stats

3D Contributions

---

Let's Connect

Open to: AI/ML Engineering • Computer Vision • GenAI/LLMOps

"Building AI systems by understanding them from first principles"