Shaownak Shahriar

• Developed an enterprise-scale vision-based workforce intelligence system for Banglalink, processing 10 concurrent IP camera streams in real time.
• Implemented face detection, recognition, and liveness detection for automated attendance, behaviour detection, and unknown-person security alerts.
• Designing scalable AI pipelines and deployment workflows with TensorRT-optimized inference for continuous production operation.

• Engineered and deployed real-time underwater image enhancement pipelines using GAN-based translation to process feeds in low-visibility environments.
• Optimized inference latency by deploying vision models onto NVIDIA Jetson Orin Nano with TensorRT, achieving stable performance for live video streams.
• Integrated Pixhawk, Raspberry Pi, and Jetson-based autonomous control systems with onboard sensors.

• Led the AI & Machine Vision sub-team; the team placed Global 1st Runner-up at RoboSub 2023 and received the Ingenuity Special Award.
• Developed underwater perception pipelines using YOLO, SSD-MobileNet, Mask R-CNN, SAM, and U2-Net with CLAHE enhancement, deployed on Jetson Orin Nano with TensorRT.
• Integrated the Pixhawk vehicle controller and VectorNav INS with the onboard AI computer and designed the onboard communication architecture.
• Collected underwater datasets and ran pool and open-water tests to validate navigation, sensing, and control.

• Led the AI & Autonomous Systems sub-team; the rover placed Global 7th at the International Rover Challenge 2023 and competed at the University Rover Challenge 2023 (Utah, USA).
• Designed real-time state estimation and localization using ROS-based SLAM and sensor fusion (Kalman/Madgwick filters) across LiDAR, GPS, IMU, and RealSense.
• Implemented inverse kinematics for a 6-DOF robotic arm and tuned closed-loop PID control for precise actuator and motor driver stabilization.
• Trained detection models (YOLO, Faster R-CNN, SSD) for terrain perception and autonomous component picking.

• Currently pursuing advanced coursework focusing on Data Science, machine learning algorithms, and advanced computing infrastructure.

• Thesis: GLE-SSD-VR: Point-Cloud-Based 3D Object Detection for Unmanned Ground Vehicles (Awarded Best Thesis of the Year 2024).
• Research Interests: 3D Object Detection, Underwater Vision Enhancement, UAV Autonomy, Imitation Learning for Robotic Control.
• Notable Courses: Quantum Computing, Artificial Intelligence, Neural Networks, Robotics, Pattern Recognition, Software Engineering, System Design.
• Mentorship: Mentored 4 undergraduate thesis groups on applied Computer Vision research (including Driver Safety, Breast Cancer Detection, and Underwater Waste Detection).
• Competitions: Represented BRAC University in multiple International Robotics Championships (UGV, AUV, UAS), national robotics competitions, and Intra-University Programming Contests (IUPC).
• Key Involvement: Co-Founder & Autonomy Lead of RaptorX, AI & Autonomous Systems Lead at Mongol-Tori, and AI & Machine Vision Team Lead at BRACU Duburi.

• Proposed a hybrid GLE-NetVR + SE-SSD voxel-point architecture for real-time LiDAR-based 3D object detection.
• Trained and benchmarked on the KITTI dataset for vehicles, pedestrians, and cyclists, achieving 88% detection accuracy.
• Implemented 3D bounding box regression with spatial attention to enhance accuracy and reduce inference latency.
• Recognized as Best Thesis of the Year 2024 by the Department of Computer Science.
• Submitted to the CVPR 2026

Built an autonomous search-and-rescue robot to assist first responders: AI-based victim detection, autonomous navigation and obstacle avoidance, and delivery of medical supplies in hazardous environments, demonstrated in simulated disaster scenarios.

Developed a mobile robot that detects objects in real time and navigates accordingly, linking a YOLOv3/OpenCV vision pipeline to Arduino-based motion control over serial UART for obstacle-aware autonomous movement.

Implemented a real-time Kalman Filter fusing MPU6050 accelerometer and gyroscope data for stable roll and pitch estimation, reducing sensor noise and drift for UAVs, mobile robots, and autonomous systems.

Built a distributed gas leak detection and automatic shut-off system: ESP32 sensor nodes publish gas readings over MQTT to a Raspberry Pi controller that closes an automated valve when a leak passes a safe threshold.

Production-grade AI platform built at Roboway Technologies for Banglalink: processes 10 concurrent IP camera streams for face recognition with liveness detection, automated attendance, behaviour detection (absence, sleep, phone use), unknown-person alerts via Telegram, and a live analytics dashboard.

Built at Dubotech: implemented FUnIE-GAN alongside CLAHE, histogram equalization, and auto color correction to restore color and contrast in low-visibility underwater footage, comparing methods qualitatively and quantitatively for AUV perception and marine inspection.

Developed a real-time driver monitoring and road perception system on Jetson Nano: drowsiness, yawning, distraction, and phone-use detection, plus lane detection and car distance via monocular depth estimation, with visual and audible alerts.

Fine-tuned YOLOv11 with DeepSORT tracking and deployed it on a Raspberry Pi 4B to detect and track UAVs from another UAV in real time, using edge-inference optimizations to maximize frame rate and minimize latency.

Developed a spatial-temporal traffic prediction framework for Dhaka's road network using Graph Neural Networks, LSTMs, and XGBoost ensembles on OSMnx road data, with interactive visualization for route optimization and logistics planning.

Designed scalable asynchronous microservices for automated CV analysis and LLM-based ranking. Built a robust NLP platform using Transformers for intelligent semantic matching, with a backend infrastructure engineered to handle asynchronous tasks and API rate limiting.

Integrated Retrieval-Augmented Generation (RAG) into a full-stack blogging platform to provide AI-driven dynamic content summarization and intelligent semantic search capabilities.

Developed a secure, end-to-end encrypted messaging platform featuring robust user authentication and cryptographic message hashing to prioritize data privacy.

Built comprehensive Enterprise Resource Planning software tailored for e-commerce operations, focusing on scalable database management and robust backend services.

Built a centralized university web application allowing students to easily report and track lost items securely across campus.

Designed and evaluated multiple deep learning architectures including LSTM, BERT, GRU, and Conv1D for advanced natural language processing and robust sentiment classification.

Engineered an accurate real estate valuation tool by fine-tuning XGBoost and custom Neural Networks to predict housing prices based on multifaceted property data.