👋

Hello, I'm Md Sazzad Hossain. Machine Learning Engineer at Devolved AI, Dhaka. Specializing in Large Language Models, Model Training, Finetuning, and Deployment.

GitHub LinkedIn Medium Kaggle HuggingFace

My experience

Machine Learning Engineer

Devolved AI - Uttara, Dhaka-BD

Responsibilities: 1. Develop, deploy, and optimize advanced machine learning models and algorithms tailored for AI-driven applications. 2. Collaborate with data science teams to analyze and preprocess large-scale datasets for training and inference. 3. Design and implement scalable machine learning pipelines to support efficient training and production-grade inference. 4. Fine-tune and enhance models to maximize performance, accuracy, and robustness. 5. Work closely with software engineering teams to seamlessly integrate AI models into production systems. 6. Conduct rigorous experiments, evaluate model performance, and iterate to achieve continuous improvements. 7. Stay informed on cutting-edge advancements in AI, including decentralized and federated learning techniques. 8. Ensure AI solutions are scalable, reliable, and optimized for efficiency. 9. Partner with cross-functional teams to identify, define, and prioritize AI-driven requirements and opportunities.

March 2024 - Current

Software Engineer

Cloud-Coder Ltd. - Gulshan-2,Dhaka,BD

Responsibilities: 1. Led the development of "RoyalGPT," a Bangla Language Model, advancing NLP for regional languages and contributing to open-source projects. 2. Developed the "SpeechToSpeech" project utilizing Bengali ASR and TTS models, integrated with the RoyalGPT Bangla Language Model. 3. Designed and implemented a Retrieval-Augmented Generation (RAG) system for RoyalGPT, enhancing its ability to retrieve and generate contextually relevant information. 4. Designed and built an ASR system for Bengali regional dialects. 5. Engineered a real-time television ad and channel detection system using detection models, optimizing performance with advanced hashing techniques for faster processing. 6. Designed and deployed a dynamic digital signage system in shopping malls, leveraging human behavior analysis to deliver personalized product information and promotions. 7. Developed a multi-camera security surveillance system for real-time tracking of human and vehicle movements, enhancing security and operational efficiency. 8. Designed a RISC-V vector extension compiler in C, optimizing performance and extending functionality for RISC-V architecture in embedded systems.

January 2022 - February 2024

My projects

AgentSync

Made with:

AgentSync is a production-ready AI agent framework designed to manage and automate complex tasks efficiently. With its Supervisor Agent module, it acts like an AI-powered manager—orchestrating multiple agents, handling workflows, and integrating with various tools like Gmail, Google Sheets, and web search utilities.

GitHub

Fattah SLM: A Bengali Small Language Model.

Made with:

This is a custom Transformers architecture designed specifically as a compact Bengali language model. For this project, I developed a custom tokenizer optimized exclusively for processing Bengali text, ensuring efficient and accurate representation of the language.

GitHub

Speech-to-Speech Chatbot

Made with:

This project is a real-time speech-to-speech chatbot that enables seamless voice-based interactions. It utilizes Automatic Speech Recognition (ASR), Voice Activity Detection (VAD), a Large Language Model (LLM), and Text-to-Speech (TTS) for a complete conversational experience.

GitHub

Background Remover Module Using U2Net.

Made with:

This repository is an easy example of background removal. I'm using U2Net model for creating musk and add some custom background. The main repo of this code is https://github.com/axinc-ai/ailia-models-tflite.