I'm a Computer Science & Engineering student from Bangladesh 🇧🇩 focused on applied AI/ML. I build LLM-powered applications, RAG systems, and computer-vision pipelines — and I document everything I learn in public. My goal is to become a research-capable AI engineer who can also ship and scale models in production.
- 🤖 Building agentic LLM apps with LangGraph, Groq, and vector databases
- 👁️ Working on object detection & tracking with YOLOv8 and ByteTrack
- 📚 Deepening my fundamentals in transformers, fine-tuning, and MLOps
- 🌱 Logging the whole journey — Python basics to agentic AI — in my-learning-hub
| Project | What it does | Built with |
|---|---|---|
| groq-langgraph-rag | RAG chatbot with a router that dynamically switches between Pinecone document retrieval and Tavily web search, orchestrated with LangGraph on Groq for low-latency responses | LangGraph Groq Pinecone Tavily |
| krishi-bondhu 🏆 | AI assistant for farmers — crop disease, pest control, and farming advice in a dual-mode RAG + web-search workflow. Built at the BUBT Tech Fusion Fest 2025 hackathon | LangGraph Groq RAG |
| IntelliReview | AI code-review platform that combines static analysis with LLMs for context-aware feedback across multi-file uploads and full repo structures | TypeScript LLMs Static Analysis |
| people-flow-tracker-yolov8 | Real-time people counting and motion tracking with entry/exit counts, line-crossing detection, and heatmaps | YOLOv8 ByteTrack OpenCV |
| qa-transformers-finetune | BERT fine-tuned on SQuAD v1.1 for extractive question answering, end to end with Hugging Face Transformers | Transformers PyTorch SQuAD |
| my-learning-hub | My public AI/ML learning journey: notes and implementations from NumPy and classical ML through CNNs, LSTMs, and agents | Python PyTorch Jupyter |
| Languages |     |
| ML / DL |      |
| LLMs & Agents |      |
| Tools |       |
My coursework and models are public too — Coursera · Hugging Face. Currently working through advanced NLP and deep learning specializations.
AI/ML internships · open-source collaboration · research opportunities
The fastest way to reach me is LinkedIn or email. If one of my projects helps you, a ⭐ goes a long way.