AI Engineer
Interactive Cares, Dhaka, Bangladesh
Jul 2025 - Present
- Developed a multi-agent AI interview platform with GPT-5 as interviewer and GPT-4o-mini for question generation, with orchestrated reasoning, structured JSON scoring, autonomous follow-up logic, and Redis-based context management, enabling <100ms responses, high concurrency, and ~60% cost optimization while preserving evaluation accuracy and safety
- Configured and optimized enterprise AI chatbot agents using LangChain for orchestration and Redis for context management, integrating GPT-4.1 via Dockerized REST APIs; implemented parallel processing with thread-pool executors and CI/CD pipelines, reducing response time by ~45% and improving throughput by ~35%
- Optimized Faster-Whisper STT and Google TTS pipelines with int8 quantization, reducing latency from ~2s to <200ms while maintaining production-grade reliability and robust debugging practices
Machine Learning Engineer
DevTechGuru Limited, Dhaka, Bangladesh
Jan 2024 - Jul 2025
- Developed and integrated an AI-assisted hip surgery system connected to C-ARM medical imaging via HDMI and Python; fine-tuned computer vision instance segmentation models for automated hip anatomy evaluation with 90%+ accuracy, reducing pre-operative planning time by ~40% while improving diagnostic efficiency under healthcare safety standards
- Engineered an end-to-end ML pipeline for vehicle plate detection using YOLOv8, achieving 93% accuracy from live CCTV feeds; deployed containerized inference via Docker with REST API integration, plus structured text extraction and automated validation workflows for reliability and scalability
- Developed backend features for a Django-based parking management system with AI-powered modules, RESTful APIs, and Redis caching for revenue dashboards and card issuance, reducing API calls by ~40% and improving data retrieval speed by ~30% for scalable production-ready delivery
ML Researcher
Health Informatics Research Lab, Daffodil International University
Dec 2022 - Dec 2023
- Published Q1 journal paper on region-of-interest embedded Graph Neural Network architecture for gallbladder cancer detection, achieving 93%+ accuracy and ~90% precision in Results in Engineering (Elsevier)
- Developed hybrid CNN-RNN architecture (NewsNet) for Bengali news categorization, achieving 94.57% accuracy and presented at IEEE conference
- Designed explainable AI framework integrating XGBoost/LightGBM with SHAP for brain tumor MRI classification, achieving 98.53% accuracy with interpretable predictions
- Engineered hybrid BERT-BiGRU model for low-resource Bengali transliteration sentence identification, advancing NLP research for regional languages
- Conducted extensive research in medical imaging, natural language processing, and deep learning applications for healthcare diagnostics