Muhammad Shurahbeel Bin Sajid — Researcher and AI Engineer

About

I’m an AI Engineer and Computer Vision researcher passionate about building intelligent systems that solve real-world problems. With a background in Mathematics and Data Science, I combine research-driven thinking with hands-on experience developing AI solutions for healthcare and human–object interaction recognition. Driven by curiosity and a commitment to continuous growth, I aspire to advance AI research and create technology with meaningful real-world impact.

As a researcher

An interaction is never contained in a single frame. Appearance tells a model what is present, motion tells it what is changing, and the meaning lies in how the two align over time. My MS thesis at Air University, supervised by Prof. Danish Hamid, turned that view into TransFlow-HOI: a transformer that fuses RGB appearance and optical-flow motion streams through cross-attention to model long-range temporal dependencies, instead of merging the streams only at the decision stage. On the HOI4D benchmark it reaches 93.1% top-1 accuracy, 14.9 points above late fusion, and is published in IEEE Access.

The same question reappears outside video. As a visiting research student at the Bioinformatics R&D Lab of Northwestern Polytechnical University, on a full scholarship, I worked on Traditional Chinese Medicine prescriptions, where the unit of meaning is a combination of herbs that a pairwise graph can only record as separate pairs. I modelled prescription records as herb–symptom networks and as hypergraphs that keep each multi-herb combination intact, then applied network-based pattern discovery to surface recurring co-prescription motifs and their symptom associations.

In industry

At CareCloud I am an AI Engineer working on clinical language systems for U.S. healthcare revenue-cycle management. I build and deploy LLM and NLP pipelines that parse and summarise clinical documents, extract structured clinical fields and feed downstream billing operations. I also develop machine-learning models for prior-authorisation and denial-management workflows, from feature engineering and evaluation design through to production integration.

I treat evaluation as the research problem inside the engineering: I define criteria with clinical and RCM domain experts and validate model outputs against their judgement. As an intern I fine-tuned open-source LLMs into a medical-document summariser that reduced clinical review time by 40% on internal evaluation sets. In open source I apply the same standard to retrieval with CrisisWeave, which validates its citations and abstains when the evidence is insufficient.

Publication

Journal article · IEEE Access · 2026 Open access

TransFlow-HOI: Transformer-Driven Multimodal Fusion for High-Performance Human–Object Interaction Recognition

Muhammad Shurahbeel Bin Sajid, Yazeed Alkhrijah, Danish Hamid, Muhammad Tahir Akram, Syed Muhammad Usman, Shehzad Khalid, Mohamad A. Alawad

IEEE Access, vol. 14, pp. 12717–12732, 2026 · doi: 10.1109/ACCESS.2026.3655585

TransFlow-HOI architecture

The TransFlow-HOI architecture fuses appearance (RGB/I3D features) and motion (optical flow) streams through cross-attention in a transformer to model long-range temporal dependencies for HOI recognition.

Top-1 accuracy on HOI4D

TransFlow-HOI (this work) 93.1%
Late Fusion 78.2%
Shallow MLP Fusion 77.3%

Read on IEEE Xplore DOI

Cite this paper BibTeX

@article{sajid2026transflowhoi,
  author  = {Bin Sajid, Muhammad Shurahbeel and Alkhrijah, Yazeed and Hamid, Danish
             and Akram, Muhammad Tahir and Usman, Syed Muhammad and Khalid, Shehzad
             and Alawad, Mohamad A.},
  title   = {{TransFlow-HOI}: Transformer-Driven Multimodal Fusion for High-Performance
             Human--Object Interaction Recognition},
  journal = {IEEE Access},
  volume  = {14},
  pages   = {12717--12732},
  year    = {2026},
  doi     = {10.1109/ACCESS.2026.3655585}
}

Experience & education

Experience

Jul 2025 — Present

AI Engineer CareCloud (MTBC), Islamabad

Clinical LLM and NLP pipelines; models for prior authorisation and denial management. Joined as AI Intern in May 2025.
Nov 2023 — Apr 2025

Graduate Research Assistant Air University, Islamabad

Designed TransFlow-HOI and built its full experimental pipeline on HOI4D.
Sep 2024 — Jan 2025

Visiting Research Student Bioinformatics R&D Lab, Northwestern Polytechnical University, Xi’an

Hypergraph modelling of Traditional Chinese Medicine prescriptions.

Education

Sep 2023 — Nov 2025

M.S. Data Science Air University, Islamabad

Thesis: Unifying Appearance and Motion: Transformer-Based Feature Fusion for HOI Recognition. Supervisor: Prof. Danish Hamid.
Feb 2016 — May 2020

B.S. Mathematics International Islamic University, Islamabad

Honours & service

2024

Outstanding Visiting Student Scholarship Northwestern Polytechnical University

Sole recipient nominated from Air University.
2025

Host and Organizer International Academic Conference on AI and Machine Learning, NWPU

Full details in my CV

Selected projects

CrisisWeave

Multimodal retrieval for disaster reporting

Evidence-first retrieval-augmented question answering over heterogeneous disaster reports: native text, OCR output, document images, video frames and transcripts, and structured tables. It validates citations, detects conflicting and stale sources, and abstains explicitly when the evidence is insufficient.

RAG
Citation validation
Guarded SQL
Python

AI Detection Suite

Real-time local inference

A local-first webcam pipeline combining YOLO object detection with CPU-based Haar face detection and lightweight hand detection. Bounded background inference and deterministic resource cleanup keep throughput stable, with an FPS and latency HUD for profiling.

YOLO
OpenCV
Python

Contact

I welcome conversations about research opportunities and collaboration in video understanding, structured representation learning and trustworthy AI.

shurahbeelsajid008 [at] gmail [dot] com

Local time in Islamabad: UTC+5