SS-2027 · rev 2026.10 · research profile · focus: clear
Hardware–software co-design for efficient machine learning on constrained devices: from the instruction set up to deployed AI systems.
AI Integration Specialist MS EE, Alabama A&M Albuquerque, NM
sharoonsharif.github.io · sharoonsharif [at] gmail [dot] com · github.com/SharoonSharif · linkedin.com/in/sharoonsharif · orcid.org/0009-0003-2669-1209 · Albuquerque, NM
Bio
I build hardware and software that let machine learning run efficiently on constrained devices. My work spans the stack: I've designed and optimized processors from the instruction set up, built FPGA accelerators for real-time video, and developed ML models for computational pathology, long-tail recognition and low-resource speech.
I hold an MS in Electrical Engineering from Alabama A&M University (4.0 GPA), where my thesis focused on AI-guided image and video processing on FPGA hardware.
Now
AI Integration Specialist, building speech and language AI for low-resource languages: production TTS and ASR, and agentic LLM systems.
Stack
Interests
Off the clock
Away from work I'm usually travelling, cooking, or reading classic literature.
Six blocks on one die. Each card matches a block in the map above.
BLK 01 · CPU · ISA
FPGA softcores for the SOMA and SORA instruction sets introduced by Jungwirth et al. (SoutheastCon 2023, 2024), taken from a single-cycle design to a four-stage pipeline, and 32-bit MIPS-based softcores with extended instruction sets.
BLK 02 · MAC ARRAY
A runtime-programmable AXI4-Stream 3×3 convolution engine for the Zynq-7020 with an analytical DDR bandwidth-feasibility model for streaming video, and a real-time PCam → Sobel → HDMI pipeline compared against Python on PYNQ. Public RTL, self-checking testbenches and post-route reports.
BLK 03 · NPU · PATH
PathQ-Former: an audit of whether histology–transcriptomics survival models actually use the transcriptome, and a one-checkpoint recipe that keeps working when a modality is missing. Five TCGA cohorts, multiple-comparison correction, and explicit reporting of where it does not beat baselines.
BLK 04 · VISION
Image–text fusion with hybrid two-stage training for 509 fungi species at a 158:1 class imbalance.
BLK 05 · DSP · SPEECH
Production TTS/ASR pipelines for low-resource languages, and a study of how orthographic convention, pseudo-labels and system combination shape code-switched Indonesian–Javanese ASR.
BLK 06 · NoC · SWARM
A pre-registered comparison of centralized planning, decentralized classical methods and multi-agent RL (MAPPO/IPPO) for multi-drone coordination, with a decentralized safety filter (2nd author).
Four papers under review: Interdisciplinary Sciences (with H. Can), TACL (with N. O’Neill), IEEE Access and IEEE TCAS-II.
Long-tail fungi classification published at the ACM Southeast Conference (ACMSE 2026).
Started as AI Integration Specialist, building speech and language AI for low-resource languages.
MS in Electrical Engineering, Alabama A&M University (GPA 4.0).
Presented two FPGA papers at IEEE AIIoT 2025.
Best Presentation Award at IEEE CCWC 2025 for the MIPS softcore paper.
Outstanding Graduate Student Award, Alabama A&M University.
Citations and the full record on Google Scholar.
Albuquerque, NM
Alabama A&M University
Machine learning and hardware systems: long-tail multimodal classification (Dr. Yujian Fu), AI-guided image/video processing on FPGA (Dr. Zhigang Xiao), and softcore processor and ISA design (Drs. Xiao, Jungwirth, Scott).
Alabama A&M University
Computer Architecture, Microprocessors, Linear Circuit Analysis, and Analog Circuits labs.
National Commission for Justice and Peace, Lahore, Pakistan
Earlier Communications Coordinator and Program Assistant. Human-rights programs, monitoring and evaluation, survey and focus-group research, and policy reports.
Micro-Nano Lab, Information Technology University of the Punjab
Thin-film deposition and spectrometry.
Alabama A&M University · GPA 4.0/4.0
Thesis: AI-guided image/video processing on FPGA. Advisor: Dr. Zhigang Xiao.
Information Technology University of the Punjab, Lahore
Advisors: Dr. Kashif Riaz, Dr. Muhammad Qasim Mehmood.