Open to opportunities
Hello, I'm

Md. Mahamudul
Hasan

I ship offline AI systems for real-world constraints — and write about what nobody tells you until it fails.

Writing

What I learned and what broke along the way.

View all →
RAG / LLMs Featured

Running RAG Completely Offline: What Nobody Tells You Until It Fails

I built two RAG systems that run entirely on CPU with no internet. Here's everything that broke — chunking strategies for equations, FAISS persistence quirks, LLM hallucination on domain-specific content, and the one config change that cut latency in half.

9 min read Read article

Projects

Mental Health Prediction App screenshot

Mental Health Prediction App

Predicts mental health risk using Random Forest and XGBoost — 85%+ prediction accuracy. Built with Streamlit for instant deployment.

Python Streamlit XGBoost
Physics Local Chatbot screenshot

Physics Local Chatbot

RAG-based physics chatbot — fully offline, no API costs. FAISS semantic search over your own PDFs; no internet, no API key, no subscription.

Python RAG NLP
Head Tracking and Blink Detection screenshot

Head Tracking & Blink Detection

OpenCV + MediaPipe system enabling hands-free mouse control for motor-disabled users — no special hardware, works with any standard webcam.

Python OpenCV MediaPipe
LocalMind AI screenshot

LocalMind AI

Private self-hosted AI platform with multi-model LLM support, voice I/O, and RAG over PDFs — real-time streaming on CPU-only hardware, zero cloud dependencies.

Python FastAPI React Native LLM
Crawl4AI screenshot

Crawl4AI

Open-source, production-ready LLM-friendly web crawler — outputs clean Markdown and JSON directly ready for LLM ingestion, no manual cleaning required.

Python Web Crawling AI/ML
KrishiBot AI Agriculture Assistant screenshot

KrishiBot — AI Agriculture Assistant

Offline-first AI agriculture assistant — identify crop diseases from a single photo, get treatment advice and weather alerts, all in bilingual English & Bengali, no agronomist needed.

Python FastAPI Next.js LLM

About me

I build AI systems that run without internet, without GPUs, and without paid APIs — based in Dhaka, Bangladesh.

I build AI systems that work where most fail: no internet, no GPU, no paid APIs. Right now that means LocalMind AI (a fully private self-hosted LLM platform with voice I/O and RAG over PDFs), KrishiBot (an offline crop disease detector for Bangladeshi farmers in English & Bengali), and Head Tracking & Blink Detection (hands-free mouse control for motor-disabled users using just a webcam).

Most of my work sits where local LLMs meet real constraints — what it actually takes to run inference on CPU, build RAG pipelines without a vector database subscription, and ship accessibility tools to users who can't afford the cloud. I work mostly in Python, FastAPI, and React Native, with OpenCV, LangChain, Ollama, and FAISS underneath.

I write about what breaks along the way — deployment failures, RAG quirks, and the gap between a working demo and a working product.

Skills

Tools I use regularly and have shipped with.

Languages

Python JavaScript Java C C++ SQL

Frameworks & Libraries

FastAPI React Native Next.js LangChain LangGraph RAG PostgreSQL Redis

AI / ML Stack

HuggingFace Google Cloud AWS Vector Databases Fine-tuning FAISS OpenCV MediaPipe Whisper

Tools & Infrastructure

Git GitHub Actions Docker Kubernetes Terraform CI/CD MLflow Linux

Experience

Timeline of roles and education

AI/ML Research & Development

Independent Research · 2022–Present

Research
  • Engineered LocalMind AI — a fully offline LLM platform with voice I/O, RAG over PDFs, and cross-platform mobile support. Zero cloud dependencies.
  • Built KrishiBot, a bilingual crop disease detector for Bangladeshi farmers running entirely on local Ollama.
  • Developed a hands-free mouse system using OpenCV + MediaPipe — no special hardware, just a webcam.
  • Shipped end-to-end ML pipelines from data collection to deployment on consumer-grade CPU hardware.
Python FastAPI Ollama OpenCV LangChain FAISS

Open Source Contributor

GitHub Community · 2021–Present

Open Source
  • Built and maintain Crawl4AI — an LLM-friendly web crawler for building clean AI training datasets.
  • Documented all projects with architecture notes, build logs, and reproducible setup guides.
  • Maintained active repositories across ML, CV, RAG, and accessibility tooling.
Python Git Markdown GitHub Actions

B.Sc. Computer Science

North South University · 2021–2025

Education
  • Built capstone-level prototypes solving real accessibility and deployment constraints.
  • Applied ML coursework directly to production projects — not just academic exercises.
  • Focused coursework on Machine Learning, Computer Vision, and NLP.
ML CV NLP Data Structures Algorithms

Higher Secondary Certificate (Science)

Shah Makhdum College, Rajshahi · 2018–2020

Education
  • Strong foundation in mathematics and problem solving that directly feeds into algorithmic thinking today.

Get in touch

Open to collaborations, research projects, and opportunities.

mahamudulh417@gmail.com @mahamudul-hasan-cse Md. Mahamudul Hasan
Dhaka, Bangladesh