Available for AI/ML roles & PhD research

Reshad Ul Karim

AI/ML Engineer & Researcher · Computer Vision · Agentic Systems · MLOps

I turn research into production AI that ships.

I build and ship explainable, edge-ready AI — multimodal computer vision, agentic LLM workflows, and the MLOps to deploy them — with measurable impact (95% fewer cloud calls, 30% faster inference).

3.87 CGPA
4 Publications
24 Citations
View Publications Download CV

About Me

Computer vision researcher building explainable, real-world AI

I am an AI/ML engineer and researcher (CGPA 3.87/4.00, BRAC University) who builds and ships multimodal computer vision, agentic LLM systems, and the MLOps to deploy them — spanning Explainable AI, robotics, edge/embedded AI, and medical imaging. My work turns research into production: a first-author IEEE Access study on explainable stroke recognition, an edge-native assistive agent that cut cloud vision calls by 95%+, ensemble road-safety models, and Mars Rover AI with up to 30% faster inference. I am open to strong PhD/research opportunities and AI/ML engineering roles where I can deliver measurable impact.

Reshad Ul Karim

Reshad Ul Karim

AI/ML Engineer & Researcher

4 Publications URC 2025 Top 8

AI/ML engineer and researcher (CGPA 3.87/4.00) with 4 peer-reviewed publications — three IEEE, including a first-author IEEE Access journal, plus JCSSE 2026 to appear. I work across computer vision, multimodal deep learning, agentic systems, robotics and MLOps — using Vision Transformers, SegFormer, YOLO, MediaPipe, LSTMs and SHAP — and I ship systems that are accurate, interpretable, and deployable at the edge (95% fewer cloud calls, 30% faster inference). Open to AI/ML engineering roles and strong PhD/research opportunities.

Download Resume Play My Journey Game
3.87 CGPA
5 Global Rank
4 Publications
6+ Languages
Google Scholar — citations ResearchGate ORCID

Technical Expertise

Hover over categories to explore my skills

AI & Machine Learning

8 Skills
90%
PyTorch
TensorFlow
YOLOv8
Computer Vision
Explainable AI

Programming

6 Languages
92%
Python
C++
JavaScript
Java
C#

Hardware & Robotics

7 Technologies
87%
Arduino
Raspberry Pi
PCB Design
ROS
Microcontrollers

Generative AI, LLMs & Agentic Systems

8 Skills
90%
Prompt Engineering & RAG
Autonomous Agents & Tool Use
Fine-tuning & RLHF
Vector Search & Embeddings
VLMs & Multimodal
MLOps (ONNX / RunPod / Guardrails)

BSc. Computer Science and Engineering

BRAC University (2022 - Present)

3.87/4.00 CGPA
Dean's List Vice Chancellor's List

Professional Experience

Building the future through innovation and research

BRAC University

Research Seed Grant Initiative 2025
Nov 2025 - Present Dhaka, Bangladesh

Undergraduate Research Assistant

Research Computer Vision Assistive Tech
DRISTEE — AI Assistive Navigation

Supporting development of DRISTEE, an AI-driven assistive navigation system for visually impaired users — contributing to technical implementation, manuscript preparation, and academic writing; integrating computer vision, depth mapping, facial recognition, cloud processing, and voice-based interaction for real-time mobility and safety assistance.

BRAC University Mars Rover Team

MONGOL TORI
Jan 2024 – Jun 2025 1.5 years

Core Team Member - AI and Autonomous Systems

AI/ML Robotics PCB Design
URC 2025 Global Finalist

Ranked Top 8 globally out of 102 teams

PCB Design & Electronics

Designed high-precision circuits for robotic arm control

AI Algorithm Optimization

Enhanced detection algorithms for autonomous systems

MONGOL TORI Phoenix 9.0
Phoenix 9.0
MONGOL TORI Hypersonic 10.0
Hypersonic 10.0
PCB Design
PCB Design
Mars Rover Team Photo
Team Photo
Download

BRAC University

Computer Science Department
Feb 2025 – Jun 2025 5 months

Undergraduate Teaching Assistant

Teaching Mentoring Academia
Student Support

Supporting 120+ students across 3 sections

Academic Assessment

Grading assignments, quizzes, and lab work

Faculty Collaboration

Working with faculty to improve course delivery

BRAC University Rescue Rover

BRACU DICHARI
Nov 2022 - Dec 2023 1 year 2 months

Contributor - Electronics and Communications

Electronics ROS Communications
Electronics Integration

Integrated basic electronics circuitry for rescue rovers

Communication Systems

Implemented rover-to-base station communication

ROS Programming

Programmed microcontrollers using ROS framework

3+

Years Experience

20+

Projects Completed

120+

Students Mentored

Top 8

Global Ranking

Research & Publications

4 publications across Explainable AI, multimodal learning, gesture HCI & medical imaging

All publications →

Research Interests

Computer Vision Multimodal Learning Agentic Systems MLOps Explainable AI (XAI) Robotics & Edge AI Medical Imaging Assistive & Autonomous Systems
Download CV GitHub LinkedIn

4

Publications

Peer-reviewed (3 IEEE + JCSSE 2026)

1st

Author, IEEE Access

Lead author on Q1 journal paper

AI/ML

Research Focus

Explainable AI & Healthcare

2024

Active Since

Continuous research output

Research Areas

Explainable AI

Developing interpretable machine learning models for healthcare applications

SHAP LIME XAI

Healthcare AI

Machine learning solutions for medical diagnosis and monitoring

Sleep Analysis Stroke Detection PPG Signals

Computer Vision

Advanced image processing and facial landmark analysis techniques

MediaPipe OpenCV Deep Learning

Publication Timeline

Conference Paper
June 24–27, 2026

Vision-based Hand Gesture Virtual Keyboard-Mouse Framework with Bilingual Next-word Prediction

Sammam Mahdi, Reshad Ul Karim, Tareq Sujat, Syeda Maliha Tabassum, Abrar Samin, Aniqua Nusrat Zereen

JCSSE 2026 — 23rd Int'l Joint Conf. on Computer Science & Software Engineering IEEE Xplore
ieeexplore.ieee.org/document/11596610

Contribution: A modular, real-time touchless keyboard-mouse that pairs geometric encoding of MediaPipe hand landmarks with two-layer LSTM bilingual (English/Bangla) next-word prediction — 9 static gestures at 97.4%+ accuracy and up to 37.7 FPS, with character error rates of 3.43% (English) and 8.34% (Bangla).

Gesture Recognition MediaPipe Bilingual LSTM Touchless HCI
Read Paper View Code
Journal Publication
March 12, 2025

Optimizing Stroke Recognition with MediaPipe and Machine Learning: An Explainable AI Approach for Facial Landmark Analysis

Karim, R. U., et al. — First author

IEEE Access Q1 Journal
doi.org/10.1109/ACCESS.2025.3550577

Contribution: Detects stroke from facial landmarks using MediaPipe and machine learning, with SHAP-based explanations so clinicians can see why each prediction is made.

Stroke Detection MediaPipe Explainable AI Healthcare
Read Paper View Code
Conference Paper
March 3, 2025

Improved Photoplethysmography-Based Four-Stage Sleep Classification with Explainable AI-Driven Machine Learning

Ferdous, T., Karim, R. U., et al.

IEEE ICEACE 2024 Dhaka, Bangladesh
doi.org/10.1109/ICEACE63551.2024.10898853
XAI Sleep Classification Research Presentation
Research Presentation Conference Talk

Contribution: Classifies four sleep stages from low-cost PPG signals using explainable ML, making interpretable sleep monitoring viable on wearable-grade sensors.

Sleep Classification XAI Healthcare PPG Signals
IEEE Xplore
Conference Paper
March 3, 2025

Machine Learning Approaches in Photoplethysmography-Based Sleep Stage Classification

Ferdous, T., Karim, R. U., et al.

IEEE ICEACE 2024 Dhaka, Bangladesh
doi.org/10.1109/ICEACE63551.2024.10898858
Machine Learning Sleep Classification Research Presentation
Research Presentation Conference Talk

Contribution: Benchmarks multiple ML algorithms for PPG-based sleep staging, establishing accuracy/efficiency trade-offs for resource-constrained, signal-only setups.

Sleep Classification Machine Learning PPG Signals Signal Processing
IEEE Xplore

Research Impact

Peer-Reviewed Publications

Three IEEE peer-reviewed papers, including a first-author study in the Q1 journal IEEE Access, spanning explainable healthcare AI and signal-based diagnostics.

Healthcare Applications

Practical implementations in healthcare settings, contributing to improved patient diagnosis and monitoring.

Open Source Contribution

Making research code and datasets available to the community for further development and validation.

Featured Projects

Research-grade systems: ViT, SegFormer, YOLO, SHAP, multimodal fusion & edge inference

Browse all projects by research domain →
Flagship Work scroll to explore
LUMENAA — edge-native assistive vision agent
Ongoing

LUMENAA — Edge-Native Assistive Vision Agent

An edge-native, agentic multimodal vision agent for visually impaired users: a planner/router with tool use and vector memory orchestrates multi-LLM/VLM reasoning (Mistral, OpenRouter) with RAG-style retrieval, a structured evaluation harness, and anti-fabrication guardrails — running perception on-device and escalating to the cloud only when needed.

0Cloud-Call Reduction
Agentic AIMultimodalRAG
View Code
WeHeal — healthcare innovation platform
Featured

WeHeal — Healthcare Innovation Platform

An AI-powered healthcare platform transforming patient care and medical diagnostics with accessible, real-time clinical tooling.

Applied AIHealthcareDiagnostics
Visit Platform
Featured

Autonomous Multi-Terrain Service Robot

A low-cost autonomous robot for indoor equipment handling with ArUco-marker navigation and robotic manipulation.

ArUco NavAutonomyManipulation
GitHub
Vision Autonomy

Mars Rover Autonomous Keyboard Typing

Vision-guided autonomous typing for Mars Rover missions — YOLOv8 keyboard detection with PaddleOCR key recognition, improving on-rover inference by up to 30%.

0Faster On-Rover Inference
YOLOv8PaddleOCRComputer Vision
GitHub
LUMENAA
Agentic AI Multimodal RAG

LUMENAA – Edge-Native Assistive Vision Agent Ongoing

An edge-native, agentic multimodal assistive vision agent for visually impaired users. A planner/router with tool use and vector memory orchestrates multi-LLM/VLM reasoning (Mistral, OpenRouter) with RAG-style retrieval, a structured evaluation harness, and anti-fabrication guardrails — running perception on-device and escalating to the cloud only when needed to achieve a 95%+ reduction in cloud vision API calls.

95%+ Cloud-Call Reduction On-Device Perception Agentic Planner + Vector Memory
Enterprise AI Document Assistant interface — notebooks, PDF upload and a cited answer
Document Assistant
RAG FastAPI Mistral

Enterprise AI Document Assistant Live demo

An AI assistant for HR policy questions that shows exactly where each answer came from — the document, the page and the quote — or admits when it doesn’t know. It also checks company policy against the law and flags the gaps.

Shows Its Sources Admits What It Doesn’t Know Runs Free, No GPU
HARP-RoPE
PyTorch Transformers BF16

HARP-RoPE – Rotary Positional Encoding for Long Contexts Research

A novel hierarchically anchored rotary positional encoding for long and structured contexts. Pre-trained 160M and 420M-parameter LLMs from scratch in PyTorch with BF16 numerical-stability analysis.

160M / 420M Pre-trained Novel RoPE Design BF16 Stability
JuktiLLM — Bangla mathematical reasoning
JuktiLLM
RLHF PyTorch Bangla NLP

JuktiLLM – RL for Bangla Reasoning Research

Applied decoupled reward–policy optimization (reinforcement learning) to improve LLM mathematical reasoning in low-resource Bangla.

Reward–Policy Decoupling RL for LLMs Low-Resource Bangla
Paraphrase Geometry
Embeddings Vector Search Retrieval

Paraphrase Geometry for Embedding Enhancement Research

A model-agnostic method that leverages paraphrase geometry to improve text embeddings and retrieval / vector-search quality.

Model-Agnostic Embedding Enhancement Retrieval Quality
CAMDet
DINOv2 Metric Depth Monocular 3D

CAMDet – Camera-Agnostic Metric 3D Detection Research

A monocular 3D object-detection framework that generalizes across cameras without retraining — DINOv2 backbone, canonical camera-space transform, and geometric-consistency metric-depth fusion, optimized for real-time edge inference.

DINOv2 Backbone Metric Depth Camera-Agnostic
Pet segmentation and breed classification
Pet Seg + Classify
Attention U-Net Segmentation Multi-Task

Pet Segmentation and Classification

A multi-task CNN on the Oxford-IIIT Pet dataset (7,349 images, 37 breeds): a shared encoder jointly performs binary foreground/background segmentation (from trimap masks) and 37-class breed classification. Base U-Net and Attention U-Net (soft attention gates on skip connections) are implemented and compared. CSE428 — Computer Vision Laboratory, Spring 2026.

Multi-Task Learning 37-Breed Classification Attention U-Net
Mars Rover Keyboard Typing Demo
Demo Video
YOLOv8 PaddleOCR Computer Vision

Mars Rover Autonomous Keyboard Typing

Autonomous keyboard typing prototype for Mars Rover missions using YOLOv8 for real-time keyboard detection and PaddleOCR for precise key recognition. Part of the vision-guided autonomy work that improved on-rover inference performance by up to 30%.

Real-time Detection Angle Mapping Rover Integration
Explainable AI sleep stage classification from photoplethysmography signals
Sleep AI Research
Machine Learning SHAP LIME

Sleep Stage Classification with XAI

Explainable AI for photoplethysmography-based sleep analysis with four-stage sleep classification and enhanced accuracy through ensemble methods.

IEEE Published Explainable AI Healthcare
Matrimonial Hub Demo
Project Demo
PHP MySQL AJAX

Matrimonial Hub - Database System

Secure matchmaking platform with advanced features, encrypted user data, dynamic search algorithms, and real-time interactions. Won 1st place in competition.

🥇 1st Place Secure Real-time
WeHeal all-in-one telemedicine and healthcare platform interface
WeHeal Healthcare
MERN Stack Telemedicine Video Consultation

WeHeal - All-in-One Healthcare Platform

Engineered an end-to-end telemedicine platform featuring OTP-secured authentication, role-based dashboards, appointment scheduling with video consultations, EMR and e-prescription management, emergency services, payment and insurance integration, online medicine and lab ordering, and feedback.

Telemedicine Video Consultation EMR System
Academic Success Prediction
Machine Learning Classification Data Analysis

Academic Success Prediction Model

Developed and evaluated multiple classification models leveraging socio-economic, demographic, and academic features to predict student outcomes as Graduate, Dropout, or Enrolled.

Predictive Analytics Student Outcomes Classification Models
Vision-based hand gesture virtual keyboard and mouse interface in use
Virtual Keyboard-Mouse
MediaPipe LSTM Bilingual

Vision-based Hand Gesture Virtual Keyboard-Mouse

Developed a modular real-time pipeline integrating MediaPipe hand-landmark gesture recognition with bilingual LSTM next-word prediction in English and Bangla within a unified keyboard-mouse interface featuring performance monitoring, delivering a low-latency, resource-efficient touchless text entry solution for desktop and wearable devices.

Hand Gesture Recognition Bilingual Prediction Touchless Interface
DRISTEE Navigation
Computer Vision NLP Cloud-based

DRISTEE – Navigation Assistant for Visually Impaired Ongoing

Developing a real-time, cloud-based navigation assistant for visually impaired users that integrates depth-aware mapping, object detection and facial recognition and voice-command NLP in a modular architecture to deliver low-latency, personalized spatial awareness and navigation guidance.

Accessibility Real-time Navigation Voice Commands
Vision Transformer-based autism spectrum disorder screening from facial and behavioral signals
Autism Detection
Vision Transformer Multimodal Fusion Healthcare

Vision Transformer-Based Autism Detection Ongoing

A Vision Transformer (ViT) framework with multimodal fusion of facial and behavioral signals for scalable, early screening of Autism Spectrum Disorder in toddlers — targeting interpretable, low-friction screening deployable outside clinical settings.

Early Detection Multimodal Analysis Healthcare AI
Arduino-based circuit schematic of the wearable health-monitoring and fall-detection device
Wearable Health Monitor
Arduino IoT Sensors GPS/GSM

Wearable Health Monitoring & Fall Detection System

Integrated wearable system combining vital sign monitoring (temperature, heart rate, SpO2, stress) with fall detection using MPU6050, GPS tracking, and real-time emergency alerts via SMS and Wi-Fi. Achieved 90% fall detection accuracy in field tests.

Elderly Care Real-time Alerts Multi-sensor Fusion
Microcontroller-free obstacle-avoiding robot
RTL Robot
RTL Logic IR Sensors L298N

Microcontroller-Free Obstacle-Avoiding Robot

An obstacle-avoiding robot with no microcontroller and no code — obstacle detection and motor control are built entirely from IR sensors, op-amp comparators and discrete transistor (RTL) logic gates driving an L298N H-bridge. A hands-on proof that autonomous behaviour doesn't always need a programmable brain. (CSE350.)

No Microcontroller Logic-Gate Control Real-time Avoidance
Autonomous Service Robot Demo
Demo Video
ESP32 ArUco Markers Computer Vision

Autonomous Multi-Terrain Equipment Service Bot

Low-cost autonomous service robot for indoor equipment handling using vision-based localization with ArUco markers. Features ESP32 microcontroller, robotic arm with pick-and-place capabilities, and multi-terrain navigation across laboratories and workshops.

ArUco Navigation Robotic Manipulation Multi-Terrain
Rakkhosh Must Die 3D OpenGL tower defense game scene
Tower Defense Game
OpenGL GLUT C++

Rakkhosh Must Die - Tower Defense Game

3D tower defense game built with OpenGL/GLUT featuring strategic defense mechanics. Defend the central rift against waves of attacking orcs using weapons, traps, turrets, and collectible powerups with increasing difficulty levels.

3D Graphics Wave System Strategic Defense
Explainable pedestrian-vehicle conflict risk prediction on a Bangladeshi road scene
Conflict Risk Prediction
SegFormer YOLOv13 SHAP

Explainable Pedestrian-Vehicle Conflict Risk Prediction Ongoing

Multi-modal fusion framework combining SegFormer road segmentation with YOLO13 detection and MediaPipe pose estimation for predicting pedestrian-vehicle conflicts in Bangladeshi road scenes. Ensemble model achieved 95.48% accuracy with SHAP-based explainability.

Traffic Safety Ensemble Learning Explainable AI

Certifications, Licenses And Courses

Verified credentials and global recognitions

AI Engineer for Data Scientists Associate

DataCamp • Issued Sep 2025

Credential ID: AEDS0010426564518

AI Fundamentals

DataCamp • Issued Sep 2025

Credential ID: AIF0025589936888

Globally Ranked 8th - University Rover Challenge 2025

The Mars Society • BRACU Mongol-Tori

Team Member, BRAC University

Honor & Awards

Recognition across research, innovation & leadership

Honor

Dean's List & Vice Chancellor's List

BRAC University

Jan 2023 – Present
1st Runner-Up

First Runner-Up

System Analysis & Design Project, Summer 2025 — School of Data & Sciences, BRAC University

2025
2nd Runner-Up

Second Runner-Up

Robosoccer, TECHSPECTRA — Robotics Club of BRAC University

2024
Grand Finalist

Third Runner-Up & Grand Finalist

EXCELERATE 2025 — Finance & Accounting Club, BRAC University

2025
Semi-Finalist

Semi-Finalist

On-Campus Rounds — Hult Prize BRAC University 2023

2023
Bronze

Bronze Award

The Duke of Edinburgh's International Award

May 2023 – Mar 2024

Cultural Leadership

Leading cultural initiatives and inspiring communities

Notre Dame Cultural Club

NDCC
Feb 2020 - Apr 2022 2+ years

General Secretary

Leadership Event Management Team Building
Team Leadership

Directed team of 30 executives and 70 volunteers

Growth Achievement

Achieved 10x audience growth with 20+ institutes collaboration

Fundraising Success

Exceeded fundraising goals by 40%

Innovation Pioneer

Pioneered "Virtual Cultural Showdown 2020" - Bangladesh's first virtual cultural contest

Cultural Performance - Singing
Performance
Event Leadership - Cultural Performance
Leadership

BRAC University Cultural Club

BuCuC
Dec 2024 – Dec 2025 1 year

Secretary - Marketing, IT, Archives & Photography

Marketing Technology Documentation
Marketing Strategy

Developing comprehensive marketing campaigns for cultural events

IT Infrastructure

Managing digital platforms and technical requirements

Documentation

Overseeing archives and photography for all cultural activities

20+

Institutes Collaborated

10x

Audience Growth

100+

Team Members Led

4+

Years Experience

Let's Connect

Open to AI/ML engineering roles, PhD/research opportunities & collaborations across CV, agentic systems & MLOps

Email

reshad.sazid@gmail.com

Location

Dhaka, Bangladesh

Image Preview

100%

Loading image...

Image preview

Research Paper

1 / 1 100%

Loading PDF...