Hi, I'm Md Shakhaout Hossain
I architect and deploy high-impact AI systems that drive business value.
As a Lead AI Engineer with 7 years of experience, I guide projects from initial concept to scalable, production-ready solutions, turning complex data into intelligent products that solve real-world challenges.
About Me
Lead AI Engineer with 7 years of software engineering experience building scalable backend platforms, LLM pipelines, and cloud infrastructure (AWS, Azure). Proven success in architecting internal tooling, standardizing MLOps workflows, and enabling product teams to deploy AI solutions efficiently.
Extensive experience designing high-performance APIs (REST, WebSocket) for real-time streaming and distributed systems. Passionate about improving developer experience (DX), mentoring cross-functional teams, and driving the production rollout of state-of-the-art AI agents across enterprise ecosystems.
Technical Skills
Backend & Platform Engineering
LLMs & AI Agents
Cloud & Infrastructure (MLOps)
Developer Tooling (DX)
Machine Learning & Computer Vision
Data Science & Analytics
Professional Experience
Sep 2022 – Present
Lead AI Engineer at Hiperdyne Corporation (Tokyo, Japan)
- Standardized AI development workflows across the organization by architecting CI/CD pipelines in Azure DevOps. Integrated UV for high-speed dependency management and Ruff/Black/Noxfile for automated code quality, significantly reducing technical debt and improving developer productivity.
- Built, operated, and scaled a fraud detection system for a telecom client using distributed AWS services (Lambda, EC2, S3, CloudWatch); maintained high reliability in production for over 3 years.
- Managed cross-functional engineering teams, mentored junior developers in technical design discussions, and collaborated directly with product stakeholders to translate high-level requirements into scalable architectural solutions.
- Led the end-to-end design and deployment of real-time, multi-threaded AI streaming applications, achieving state-of-the-art performance and sub-33 ms latency in production environments.
- Designed a privacy-preserving computer vision application using diffusion models; awarded Japanese patent (特許7498534).
Aug 2019 – Aug 2022
Machine Learning Engineer at Hiperdyne Corporation (Tokyo, Japan)
- Engineered and deployed computer vision models for automated quality inspection, reducing manufacturing defects by an estimated 15%.
- Developed a high-throughput fraud detection system for a large-scale consumer application, processing millions of transactions daily.
Nov 2017 – Dec 2018
Research Assistant – LGED Project (Dhaka, Bangladesh)
- Contributed to mechanical and electrical design, procurement, and installation of automated systems.
Featured Projects
Leadership Role • Real-time Systems
Real-Time Active Speaker Detection & Streaming Pipeline
Led the end-to-end development of an Active Speaker Detection (ASD) system utilizing a bidirectional WebSocket architecture for low-latency, full-duplex client-server communication. Architected a streamlined pipeline integrating voice embedding extraction and Azure APIs.
Impact:
Drove the engineering lifecycle via Agile/Scrum methodologies and established automated CI/CD pipelines to ensure highly structured, reliable production deployments.
Leadership Role • Computer Vision
Real-time Multi-Camera Multi-Person Tracking System
As project lead, I directed the design and execution of a system optimized for high-load inference. Solved distributed systems challenges by implementing multi-threading, multi-processing, and a shared ring buffer with semaphore control to prevent race conditions and out-of-memory errors
Impact:
Reduced memory usage by 20% via optimized SQL queries and deployed the entire architecture using Docker, achieving scalable state-of-the-art tracking performance.
Tech Stack
Published Patent • Generative AI
Privacy-Preserving Face Swapping with Diffusion Models
Researched and implemented LoRA-based fine-tuning of Stable Diffusion for high-precision face swapping in 2D videos. Developed a privacy-preserving application that anonymizes individuals while maintaining scene realism.
Impact:
Resulted in a Japanese patent, showcasing innovation in ethical AI and image processing.
Edge AI • Computer Vision
Real-time Object Pose Estimation on Edge Devices
Trained HRNet model for object keypoint detection using COCO Annotator. Deployed on NVIDIA Jetson Xavier NX achieving 25 FPS inference speed on live webcam feeds.
Impact:
Enabled real-time pose estimation for industrial automation use cases.
Tech Stack
Computer Vision
Automated Parking Lot Management System
Managed the end-to-end data pipeline, from quadcopter image acquisition to training a YOLOv3 model for automated vehicle detection.
Impact:
The system provided the foundational technology for a real-time parking availability service, automating a previously manual process.
MLOps • Cloud Automation
AWS Fraud Detection Infrastructure
Owned the end-to-end design and production rollout of a fraud anomaly detection platform using AWS (Lambda, EC2, S3, SNS).
Impact:
Built automated infrastructure workflows and alerting systems that reduced manual investigation time by 70%.
Edge AI • Real-time Systems
Low-Latency Streaming Application
Built a real-time, multi-channel streaming app on NVIDIA Jetson, synchronizing video and audio with ultra-low latency using UDP for a telemedicine use-case.
Impact:
Created a responsive prototype for a major manufacturer, proving the feasibility of complex, low-latency streaming on embedded hardware.
Tech Stack
Predictive Maintenance
Mechanical Valve Leakage Classification
Developed classification models using XGBoost, Random Forest, and deep learning to differentiate leakage rates in mechanical valves.
Impact:
Achieved 98% accuracy with PCA-enhanced feature engineering. Delivered a production-ready software package to the client with full documentation and testing.
Tech Stack
Predictive Maintenance
Predictive Maintenance for Railway Infrastructure
Developed regression models from large-scale sensor data to predict potential structural failures in railway tracks.
Impact:
Created a predictive model to help maintenance teams proactively address points of failure, enhancing safety and operational efficiency.
Tech Stack
Unsupervised Learning
Solar Panel Anomaly Detection
Trained an AutoEncoder model to automatically detect defects in solar panels from aerial video, developing the full data pipeline.
Impact:
Delivered an automated inspection system that significantly improved maintenance efficiency and reduced inspection costs for the solar plant.
Tech Stack
NLP • Generative AI
NLP Prompt Engineering for LLM Reliability
Researched and applied advanced prompting techniques (e.g., CoT, ReAct) to improve the reliability and task-specificity of Large Language Models.
Impact:
Built a demonstration system that significantly suppressed LLM hallucinations and steered model behavior towards precise, verifiable outcomes.
Tech Stack
Research Implementation
Human Biomechanics Estimation from 2D Video
Implemented biomechanics-based model using HRNet to estimate human COM and torque at body segments from 2D videos. Applied cadaver-derived formulas for accurate physical modeling.
Impact:
Supporting ergonomic risk analysis and posture evaluation.
Education
2022 - current
PhD candidate in Information Science
Nara Institute of Science and Technology, Nara, Japan
Research: Deep learning for medical imaging, blur detection, and image denoising using diffusion models.
Sep 2017
Bachelor of Science in Mechanical Engineering
Bangladesh University of Engineering and Technology, Dhaka, Bangladesh
Thesis: STUDY ON HEAT TRANSFER CHARACTERISTICS OF CLOSED LOOP HEAT PIPE FOR DIFFERENT WORKING FLUIDS WITH INTERNAL FIN
Publications & Patents
📜 Publications
No-Reference Blurred Image Detection from Colonoscopy Videos Using Walsh-Hadamard Transform and Kolmogorov Smirnov Test
MD Shakhaout Hossain, Naoaki Ono, Shigehiko Kanaya, Md. Altaf-Ul-Amin, 2024. Published in IEEE International Workshop on Imaging Systems and Techniques (IST).
View PublicationErgonomic Risk Prediction for Awkward Postures From 3D Keypoints Using Deep Learning
Md Shakhaout Hossain, Sami Azam, Asif Karim, Sidratul Montaha, Ryana Quadir, Friso De Boer, 2023. Published in IEEE Access.
View PublicationAutomatic Rail Crossing Alarming System
Sums Uz Zaman, Shakhaout Hossain, Celia Shahnaz, 2019. Published in 2018 IEEE International WIE Conference on Electrical and Computer Engineering (WIECON-ECE).
View Publication💡 Patents
コンピュータプログラム、画像処理方法及び画像処理装置
Inventors: ホッサイン シャカウト, 五十嵐 一浩, 山下 真吾, 宮島 靖, アブドスサブル ムハマッド. Patent No. P7498534. Issued on 2024.6.12.
View Patent DetailsCertifications & Professional Development
Machine Learning and Deep Learning with Python
Hiperdyne
Fundamentals of Reinforcement Learning
Coursera
Short Course on Nuclear Power Engineering
INPE
Extracurricular Activities
![[Your University Robotics Club] team photo](./images/techkriti.jpg)
BUET Fireflies
Robotics Team Lead
Led a team of 3 students in designing and building an autonomous Maze solver robot for an international competition, securing champion. Responsible for programming, mechanical design and sensor integration.
![[Your University Robotics Club] team photo](./images/techfest.jpg)
BUET Fireflies
Robotics Team Member
Secured Top 5 position in an international Maze solver and Grab&Pick robotics competition. Responsible for mechanical design and sensor integration.
![[Your University Robotics Club] team photo](./images/ERC.jpg)
Interplanetars
Robotics Team Member
Participated in European Rover Challenge (ERC) competition in Poland. Responsible for mechanical design and sensor integration.
![[Your University Robotics Club] team photo](./images/BRS.jpg)
BUET Robotics Society (BRS)
Co-Founder & General Secretary
Beyond my professional work, I was one of the founding members of my university's robotics club. I have arranged and managed national level robotics competition and robotics workshops.
![[Your University Robotics Club] team photo](./images/ecorun.jpg)
BUET TEKTON
Team Member
As a founding member of BUET Tekton, we participated in fuel efficient vehicle design competition. I was responsible for mechanical design, procurement and team management.
Hobbies & Interests
Hiking & The Outdoors
I find clarity and challenge on the trail. Hiking is my way to disconnect from technology and reconnect with nature. It's taught me the value of persistence, preparation, and appreciating the journey as much as the destination.
Moments from the Trail
Mount Fuji, Japan (highest mountain of Japan)
Mount Kumotori, Tokyo, Japan (highest peak of Tokyo)
Mount Siraiwayama, Saitama, Japan
Cultural Exploration & Travel
Traveling opens my mind to new perspectives and ways of life. I enjoy immersing myself in nature, trying new foods, and understanding history firsthand. This curiosity and adaptability translate directly into my work.
The Travel Log
Osaka castle, Osaka, Japan
Kegon falls, Nikko, Japan
Lake Ashi, Hakone, Japan
Awards
Champion
International Autonomous Robotics Challenge, IIT Kanpur, India, Techkriti-2015
2nd Runner-up
Formula R Robotics Competition organized by Techfiesta’14, BUET
Language Proficiency
Bengali
Native
English
Fluent
Japanese
Basic (N5)
Let's Connect
I'm currently open to new opportunities and collaborations. If you have a project in mind or just want to chat about AI, feel free to reach out.