Mehedi H.

I build AI systems that actually get used in Production. Machine Learning Engineer with 4+ years of experience building end-to-end AI solutions, mobile and backend systems, and deploying models to production at scale. Computer vision (CV) is the strongest part of my background. I have extensive experience in deploying CV models using CCTV camera footage and streams. I developed the largest CV pipeline in Bangladesh, capable of processing approximately 1.1 GB/s of CCTV camera streams using C++. Additionally, I have created and optimized machine learning models for real-world applications, including converting them to ONNX and TensorRT formats to ensure efficient performance in constrained environments. My background includes working with technologies such as RTSP, GStreamer, Redis, Apache Kafka, Apache Flink, and more to build large-scale video streaming pipelines. The largest CV model optimization feat to date is that I developed and optimized a model on a low-end CPU (Intel Xeon 4309Y with RTX 4070), increasing the frame rate from 12 FPS to ~210 FPS. I've recently focused on Generative AI (GenAI), developing AI agents for workflows and customer challenges. My projects include data extraction, support roles, and chatbots, all operational in production and providing consistent results. I've found that smaller, specialized models are more effective for specific tasks than using one large model. I tend to focus on simple, clean solutions that work well in practice — not overengineered systems that are hard to maintain. I am a problem solver by nature. My tech stack is the tools I mostly work with, but not limited to. Tech Stack: - Core: Python, C++, PyTorch, FastAPI - Data & Infra: MongoDB, Docker, Redis, Apache Kafka, Apache Airflow - Systems: Distributed Pipelines, Concurrency, ETL Orchestration, Event-Driven Systems - Specialization: AI Agents, LLM Apps, Real-Time Vision Inference, Vision Optimization