Abdulsamad Ibrahim
Deep learning research engineer with hands-on experience designing, training, and optimizing multimodal AI models
Hire this person
Send a job offer directly to this candidate
About
Deep learning research engineer with hands-on experience designing, training, and optimizing multimodal AI models including a CNN+Transformer hybrid architecture alongside real-time data-acquisition systems and low-latency signal processing (C++, MATLAB, kernel-level fault diagnosis). Proficient in PyTorch, TensorFlow, and the full Python ML pipeline, with practical Linux/Ubuntu systems experience and Docker containerization. Seeking an ML Engineer, Algorithm Engineer, or Data Mining role to build proven model-design and optimization skills toward larger-scale ML infrastructure and inference systems.
Experience
Research Assistant Intern - Neural Dynamics & Experimental Data Systems at Lab of Neural Dynamics of Visual Perception and Cognition, ICPBR, CEBSIT, CAS, Shanghai (2026-04 – Present)
- Set up NI instrument configuration and MonkeyLogic integration on Windows and Ubuntu for neural data acquisition.
- Optimized an iSpeed camera data-transmission MATLAB pipeline, reducing latency from 48ms to 16ms, then implemented a C++ receiver-side solution reaching 4ms (250Hz).
- Built a GPIO-triggered real-time video synchronization system (≤12ms trigger-to-recording latency) using an NI PCIe-6321 digital I/O card, diagnosing and resolving a kernel-level driver fault during a hardware migration from Raspberry Pi to a DAQ-based architecture.
- Developed and maintained Python/MATLAB scripts for experimental control, data preprocessing, and statistical analysis; managed research-computing infrastructure including backup and storage-reliability protocols.
Graduate Researcher — Deep Learning for Multimodal Gesture Recognition at Embedded Systems and IoT Research Group, School of Computer Science and Technology, USTC (2024-10 – 2026-06)
- Designed and trained a Hybrid SqueezeFormer (CNN+Transformer) architecture and a pure Transformer architecture for 250-class isolated sign-language classification from video-derived landmark features; the production-optimized variant reached ~90% top-1 / 98% top-5 accuracy at 922K parameters — roughly 3x smaller and 2x faster to train than the 2.7M-parameter baseline (85.12% top-1 / 96.88% top-5).
- Developed a complete inference pipeline from raw video through landmark extraction, temporal standardization, classification, sentence reconstruction, and LLM-based refinement into English speech output.
System Support Engineer at Dufil Prima Foods Limited (2023-03 – 2023-08)
- Achieved an 80% user-experience rating across departments and employee levels (expatriates and regular employees).
- Collaborated with an AI team (India) on an accident-prevention system, supplying real-world CCTV recordings used to train the model.
- Installed and configured Sophos routers and Ubiquiti/TP-Link switches; managed VMware virtual servers with monthly backup strategies sustaining 99% data availability.
Education
M.Sc. in Computer Technology – University of Science and Technology of China (2023-09 – 2026-06)
B.Eng. in Electrical and Electronics Engineering – Kebbi State University of Science and Technology (2015-10 – 2021-08)
