Open to opportunities

SAMMAM MAHDI

@sammammahdi

Message

I’m a final-year AI researcher focused on efficient, trustworthy deep learning via distillation and model compression.

Bangladesh

Message

What I'm looking for

I’m looking for an applied research role where I can publish at top-tier venues and also ship research into real systems—especially work on efficient, trustworthy deep learning with low-latency multimodal pipelines and deployable model compression.

I’m a final-year Computer Science researcher working on efficient, trustworthy deep learning. I use knowledge distillation, model compression, and representation learning to make large models smaller, faster, and more deployable without sacrificing quality.

My research has produced peer-reviewed results in IEEE Access and IEEE ICEACE, with total citations reported as 21 on Google Scholar (May 2026). For example, I engineered an explainable AI pipeline using feature importance over 228 MediaPipe facial landmarks and co-designed a Multimodal Voting Classifier to achieve 94.75% diagnostic accuracy for stroke/non-stroke facial image classification.

Right now, I’m expanding that focus across distillation/compression for federated quantum-classical attention models, low-latency multimodal vision/VLM pipelines for real-time inference, and state-space / non-Transformer architectures (Mamba-2). I enjoy taking ideas from paper → code → production, and I’m seeking an applied-research role where I can publish at top-tier venues (NeurIPS, ICML, ICLR, COLM) while shipping research into real systems.

Experience

Work history, roles, and key accomplishments

Current

ICEA Deepfake Forensics

Current

BRAC University

Aug 2025 - Present (10 months)

Designed ICEA, a physics-grounded deepfake detection framework that flags violations of invariant laws to reduce reliance on generator-specific artifacts and improve cross-generator generalization. Built a dual-encoder, evidence-accumulating classifier that produces multi-label, human-interpretable reasons for each deepfake verdict and is engineered for distribution-shift robustness.

Deepfake Detection Representation Learning Dual Encoder Architectures Multi Label Classification Distribution Shift Robustness

Current

Quantum-Classical Distillation

Current

BRAC University

Jan 2025 - Present (1 year 5 months)

Applied knowledge distillation to compress quantum-classical hybrid attention models under strict parameter and communication budgets. Investigated privacy-preserving quantum federated learning using distillation for data-efficient, compute-constrained training.

Knowledge Distillation Model Compression Federated Learning Parameter Efficient Attention Models PyTorch Transformers PQC (Parameterized Quantum Circuits)Representation Learning

Current

DRISTEE Agentic Vision System

Current

BRAC University

Jan 2025 - Present (1 year 5 months)

Architected a modular agentic assistive vision pipeline for the visually impaired with real-time, inference-constrained operation behind a single low-latency speech interface. Built depth-based 3D perception using CAMDet and MASt3R-SLAM mapping, integrated multimodal VLM-based OCR and grounded scene Q&A, and engineered semantic caching to meet tight latency/memory budgets.

SLAM (MASt3R SLAM)RetinaFace

Current

Teaching Assistant (CSE)

Current

BRAC University

Jan 2025 - Present (1 year 5 months)

Mentored 100+ undergraduate students by leading weekly recitations on algorithm complexity, graph algorithms, and dynamic programming. Supported student learning through structured practice and problem-solving sessions.

Algorithmic Complexity Dynamic Programming Mentorship Recitation Facilitation

Mars Rover AI/ML Contributor

BRAC University

Jan 2023 - Jan 2025 (2 years)

Contributed to the University Rover Challenge team by designing and training YOLOv8 models for real-time, on-device object detection under outdoor lighting variability. Optimized models for field inference on edge hardware for bottles, mallets, and keyboard keys.

YOLOv8 Object Detection On Device Inference Training Under Lighting Variability Computer Vision Python Linux

Education

Degrees, certifications, and relevant coursework

BRAC University

Bachelor of Science, Computer Science and Engineering

Grade: CGPA: 3.92/4.00; Highest Distinction

Pursuing a Bachelor of Science in Computer Science and Engineering at BRAC University, expected to complete in May 2026. Earned a CGPA of 3.92/4.00 with Highest Distinction and multiple Dean’s List awards.