I care more about understanding how things work than just making them work. That started back in undergrad at the University of Rajshahi, where I spent a lot of late nights doing computer vision research for medical imaging. I wasn't just trying to get a model to work, I wanted to know why it made the decisions it did. That same curiosity eventually pulled me toward Large Language Models and AI agents, and lately I've been thinking about how these systems might move past the cloud and into the physical world through IoT, edge devices, and robotics.

Day to day, I turn research ideas into things people actually use. That means building ML/LLM pipelines along with the deployment infrastructure, MLOps, and DevOps practices that keep them running reliably in production. Still, the part I enjoy most is the research itself: getting stuck on an idea for days and then finally watching it click.

Feel free to reach out and connect!

📖 Currently: AI Engineer at Vivasoft Limited, building and deploying production ML/LLM systems

📍 Research Focus: Efficient, interpretable AI systems, including LLM/agent inference optimization, MLOps and DevOps for ML infrastructure, and edge/IoT/robotics deployment

🎓 Next Chapter: Working toward a PhD in AI systems research, focused on efficiency, interpretability, and real world robustness beyond the benchmark

• Geo-HySTE: Geometric hybrid spatio-temporal encoder for enhanced 3D human pose estimation

  Sanjoy Kumar, Sangeeta Biswas, Md Ahanaf Arif Khan, Md Hasnain Ali, Mehedi Hasan Shakil
  Computers and Electrical Engineering, Vol. 139, Article 111375, November 2026

• Interpreting Deep Neural Networks in Diabetic Retinopathy Grading: A Comparison with Human Decision Criteria

  Sangeeta Biswas, Md. Ahanaf Arif Khan, Md. Hasnain Ali, Johan Rohdin, Subrata Pramanik, Md. Iqbal Aziz Khan, Sanjoy Kumar Chakravarty, Bimal Kumar Pramanik
  Life, Vol. 15, No. 9, Article 1473, September 2025

• Competency Comparison of Deep Neural Networks for Identifying Gender in Color Fundus Photographs

  Md. Ahanaf Arif Khan, Md. Hasnain Ali, Nirjor Saha, Md. Sadman Shakib Shoumik, Sangeeta Biswas
  2023 26th International Conference on Computer and Information Technology (ICCIT), 13-15 December 2023, Cox's Bazar, Bangladesh

🥇 Winner – AI Hackathon 2025 Developed VisionDesk, an AI-driven surveillance solution for manufacturing and industrial environments. Leveraged advanced pose estimation and activity recognition to distinguish productive behaviors (e.g., assembling, typing) from non-productive actions (e.g., idling, phone usage). Provided actionable insights to optimize workforce efficiency, process automation, quality control, and supply chain performance.