Omid Nezami

Omid Nezami

Principal Applied Scientist
Oracle · Sydney, Australia

Google Scholar LinkedIn GitHub

Research & applied science

I work on large language models, clinical natural language processing and evidence-grounded AI agents.

As a Principal Applied Scientist at Oracle, I combine model development with coordination of clinical coding-agent evaluation, error analysis and release readiness. My research spans preference alignment training, data synthesis, multimodal learning and evolutionary optimisation.

I completed my PhD in Computer Science at Macquarie University in 2019, supervised by Professor Mark Dras, followed by postdoctoral research, machine learning teaching and postgraduate project advising at Macquarie.

Research

Themes &
contributions

Large language models & data-centric NLP

CodeLLM data synthesis, preference alignment, knowledge-augmented fine-tuning, model adaptation and long-text processing.

Evidence-grounded clinical coding agents

Diagnosis and procedure extraction, source-linked justifications, temporal reasoning, rule-guided validation and human review, supported by reproducible end-to-end evaluation.

Healthcare NLP & privacy

Clinical named entity recognition, PII/PHI de-identification, medical semantic textual similarity, training-data augmentation and interdisciplinary psychopathology research.

Multimodal learning & robust perception

Controlled image captioning, facial expression analysis, speech emotion, student engagement recognition and adversarial object-detection attacks.

Evolutionary optimisation & robotics

Diversity-guided particle swarm optimisation, including DDEPSO and DGPSO, and optimisation of biped locomotion.

Experience

Research to
real-world systems

2021–present

Principal Applied Scientist

Oracle · Sydney, Australia

  • Develop and evaluate LLM-based clinical coding agents, including evidence-linked diagnosis and procedure outputs, workflow refinement and systematic error analysis.
  • Coordinate model-improvement priorities, code reviews, release-readiness follow-up and customer feedback with engineering and clinical coding specialists.
  • Develop end-to-end and component-level evaluation workflows, regression checks and comparisons of model quality, cost and latency.
  • Led LLM preference-alignment development for healthcare; initiated, developed and released a PII/PHI de-identification service supporting medical-data acquisition.
  • Contributed to custom and pre-built clinical NER and co-invented adaptive and gazetteer/perturbation-based training-data augmentation.
  • Co-invented preference-aligned form conversion, multi-head fine-tuning and long-text methods, with published applications addressing knowledge augmentation and low-resource code alignment.
  • Co-authored the NAACL 2025 CodeLLM data-synthesis study; developed machine learning software using Python, TensorFlow and PyTorch.

2020–2021

Postdoctoral Research Fellow

Macquarie University · Sydney, Australia

Researched deep learning, transformer architectures and transfer learning for healthcare NLP, including semantic textual similarity in the medical domain. Developed and evaluated models using Python, TensorFlow, pandas and PySpark, supervised by Professor Mark Dras.

2018–2019

Research Assistant

Macquarie University · Sydney, Australia

Applied deep learning, transfer learning and data augmentation to computer vision and classification in collaboration with Western Sydney University and the University of Wollongong. Developed models using Python, TensorFlow and MATLAB, supervised by Professor Len Hamey.

Publications

Selected work

12 publications, newest first.

Full list on Google Scholar ↗

CV snapshot, 21 September 2026: 790 citations · h-index 12 · i10-index 12.

  1. 2025 · NAACL-HLT

    Mastering the Craft of Data Synthesis for CodeLLMs

    Meng Chen, Philip Arthur, Qianyu Feng, Cong Duy Vu Hoang, Yu-Heng Hong, Mahdi Kazemi Moghaddam, Omid Nezami, Duc Thien Nguyen, Gioacchino Tangari, Duy Vu, Thanh Vu, Mark Johnson, Krishnaram Kenthapadi, Don Dharmasiri, Long Duong, and Yuan-Fang Li.

    Proceedings of NAACL-HLT 2025, Volume 1: Long Papers, pp. 12484–12500. Albuquerque, New Mexico, USA.

    PaperDOI

  2. 2024 · Psychological Medicine

    Elemental Psychopathology: Distilling Constituent Symptoms and Patterns of Repetition in the Diagnostic Criteria of the DSM-5

    Miriam K. Forbes, Bryan Neo, Omid Nezami, Eiko I. Fried, Katherine Faure, Brier Michelsen, Maddison Twose, and Mark Dras.

    Psychological Medicine, 54(5):886–894.

  3. 2021 · Computer Vision and Image Understanding

    Pick-Object-Attack: Type-Specific Adversarial Attack for Object Detection

    Omid Nezami, Akshay Chaturvedi, Mark Dras, and Utpal Garain.

    Computer Vision and Image Understanding, 211:103257.

    PaperPDFCode

  4. 2020 · Journal of Artificial Intelligence Research

    Image Captioning using Facial Expression and Attention

    Omid Nezami, Mark Dras, Stephen Wan, and Cécile Paris.

    Journal of Artificial Intelligence Research, 68:661–689.

    PaperPDF

  5. 2019 · ECML-PKDD

    Automatic Recognition of Student Engagement using Deep Learning and Facial Expression

    Omid Nezami, Mark Dras, Len Hamey, Deborah Richards, Stephen Wan, and Cécile Paris.

    Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pp. 273–289. Würzburg, Germany.

    PaperPDFCode

  6. 2019 · Language Resources and Evaluation

    ShEMO: A Large-Scale Validated Database for Persian Speech Emotion Detection

    Omid Nezami, Paria Jamshid Lou, and Mansoureh Karami.

    Language Resources and Evaluation, 53(1):1–16.

    PaperPDFData

  7. 2019 · PRICAI

    Towards Generating Stylized Image Captions via Adversarial Training

    Omid Nezami, Mark Dras, Stephen Wan, Cécile Paris, and Len Hamey.

    Pacific Rim International Conference on Artificial Intelligence, pp. 270–284. Cuvu, Fiji.

    PaperPDF

  8. 2018 · ECML-PKDD

    Face-Cap: Image Captioning using Facial Expression Analysis

    Omid Nezami, Mark Dras, Peter Anderson, and Len Hamey.

    Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pp. 226–240. Dublin, Ireland.

    PaperPDFData

  9. 2017 · PACIS

    Semi-Supervised Detection of Student Engagement

    Omid Nezami, Deborah Richards, and Len Hamey.

    21st Pacific-Asia Conference on Information Systems. Langkawi, Malaysia.

  10. 2013 · Procedia Computer Science

    Dynamic Diversity Enhancement in Particle Swarm Optimization (DDEPSO) Algorithm for Preventing from Premature Convergence

    Omid Nezami, Anvar Bahrampour, and Paria Jamshidlou.

    Procedia Computer Science, 24:54–65.

    Paper

  11. 2013 · ICAI

    Diversity Guided Particle Swarm Optimization Algorithm based on Search Space Awareness Particle Dispersion (DGPSO)

    Anvar Bahrampour and Omid Nezami.

    Proceedings of the 2013 International Conference on Artificial Intelligence, pp. 218–224. Las Vegas, USA: CSREA Press.

    PDF

  12. 2010 · RoboCup 2009

    Evolution of Biped Walking Using Truncated Fourier Series and Particle Swarm Optimization

    Nima Shafii, Siavash Aslani, Omid Nezami, and Saeed Shiry.

    RoboCup 2009: Robot Soccer World Cup XIII, Lecture Notes in Computer Science, vol. 5949, pp. 344–354. Springer.

    Paper / DOI

Patents

Research
translation

Co-inventor across seven patent families: five granted US patents and two published US applications.

Assignee: Oracle International Corporation. Statuses as at 21 September 2026. Ordered by grant date or application publication date.

  1. Published application

    Knowledge augmentation in supervised instruction fine-tuning of generative models

    US 2026/0259709 A1 · 3 September 2026

    Filed 4 February 2026 · Application 19/529,887

    Inventors

    Duc Thien Nguyen, Tuyen Quang Pham, Prashan Madumal, Omid Nezami, Mohammad Mahdi Kazemi Moghaddam, Thanh Tien Vu, Yakupitiyage Don Thanuja Samodhye Dharmasiri, and Thanh Long Duong.

    Patent record ↗
  2. Granted US patent

    Form conversion using preference alignment training

    US 12,694,019 B1 · 28 July 2026

    Application 19/040,650

    Inventors

    Thanh Tien Vu, Prashan Madumal, Omid Nezami, Cong Duy Vu Hoang, Gioacchino Tangari, Duy Vu, Dai Quoc Nguyen, Yakupitiyage Don Thanuja Samodhye Dharmasiri, Ying Xu, Thanh Long Duong, Clemence Andree Simone Lanfranchi, Damien Alexandre Hilloulin, Rhicheek Patra, Sungpack Hong, and Hassan Chafi.

    Patent record ↗
  3. Published application

    Correction based preference alignment fine-tuning for low resource programming languages

    US 2026/0212265 A1 · 23 July 2026

    Filed 12 June 2025 · Application 19/236,055

    Inventors

    Prashan Madumal, Duc Thien Nguyen, Omid Nezami, Dai Quoc Nguyen, Mohammad Mahdi Kazemi Moghaddam, Philip Arthur, Tuyen Quang Pham, Thanh Tien Vu, Cong Duy Vu Hoang, Yakupitiyage Don Thanuja Samodhye Dharmasiri, and Thanh Long Duong.

    Patent record ↗
  4. Granted US patent

    Fine-tuning multi-head network from a single transformer layer of pre-trained language model

    US 12,512,091 B2 · 30 December 2025

    Related published application: US 2026/0080864 A1 · 19 March 2026 · Application 19/402,418

    Inventors

    Thanh Tien Vu, Tuyen Quang Pham, Omid Nezami, Mark Edward Johnson, Thanh Long Duong, and Cong Duy Vu Hoang.

  5. Granted US patent

    Adaptive training data augmentation to facilitate training named entity recognition models

    US 12,499,385 B2 · 16 December 2025

    Related published application: US 2026/0065171 A1 · 5 March 2026 · Application 19/385,550

    Inventors

    Omid Nezami, Thanh Tien Vu, Budhaditya Saha, and Shubham Pawankumar Shah.

  6. Granted US patent

    Training data augmentation using gazetteers and perturbations to facilitate training named entity recognition models

    US 12,417,348 B2 · 16 September 2025

    Application publication preceding grant: US 2023/0325599 A1 · 12 October 2023 · Application 18/185,675

    Related published application: US 2025/0371269 A1 · 4 December 2025 · Application 19/303,714

    Inventors

    Omid Nezami, Shivashankar Subramanian, Thanh Tien Vu, Tuyen Quang Pham, Budhaditya Saha, Aashna Devang Kanuga, and Shubham Pawankumar Shah.

  7. Granted US patent

    System and techniques for handling long text for pre-trained language models

    US 12,210,830 B2 · 28 January 2025

    Related published application: US 2025/0117585 A1 · 10 April 2025 · Application 18/987,825

    Inventors

    Thanh Tien Vu, Tuyen Quang Pham, Mark Edward Johnson, Thanh Long Duong, Ying Xu, Poorya Zaremoodi, Omid Nezami, Budhaditya Saha, and Cong Duy Vu Hoang.

Teaching

Learning through
research & practice

2019–2020

Sessional Teaching Academic

Faculty of Science and Engineering, Macquarie University

Taught machine learning, data science and research computing, designing and guiding project-based assignments using deep learning and conventional methods. Worked with unit convenor Professor Mark Dras.

  • COMP8220 — Machine Learning
  • COMP777 — Computing Methods for Research
  • COMP783 — Data Science and Machine Learning

2020–2021

Postgraduate internship project advising

Macquarie University

Advised postgraduate internship projects on Data Mining for Route Optimization Problem.

Education

Academic
background

July 2016–December 2019

PhD in Computer Science

Macquarie University, Australia

Thesis: Explorations in Controlled Image Captioning.
Advisor: Professor Mark Dras.

Supported by MQRES and a CSIRO Data61 PhD top-up scholarship.

September 2008–March 2011

MSc in Computer Engineering — Artificial Intelligence

Qazvin University, Iran · First Class Honours

Thesis: Biped Walking Robots using Truncated Fourier Series and Optimization Algorithms.

September 2004–July 2008

BSc in Computer Engineering

Qazvin University, Iran · First Class Honours

Recognition & service

Awards, support
& community

Awards & recognition

  • Oracle Innovation Prize (Edison Award) — Sydney, 2023.
  • First place, Industry Event Poster Competition — Macquarie University, 2017.
  • RoboCup Mixed Reality League — third place, Singapore, 2010; fourth place, Graz, Austria, 2009.
  • RoboCup Iran Open — first place, Mixed Reality League and Technical Challenge, Tehran, 2010; first place, Mixed Reality League, Qazvin, 2009.

Scholarships & research support

  • CSIRO Data61 PhD Top-Up Scholarship, 2018–2019.
  • Macquarie University Postgraduate Research Fund (PGRF), 2018.
  • Macquarie University Research Excellence Scholarship (MQRES), 2016–2019.

Professional service

Conference reviewing, 2019–2021: AAAI (2020, 2021), ACL (2021), EMNLP (2019, 2020), COLING (2020), EACL (2021) and NAACL (2021).

Technical Committee Member, RoboCup, Iran, 2011–2014: organised robotics competitions at national and international levels.

Contact

Get in touch

Sydney, Australia

Find my full publication record on Google Scholar, connect on LinkedIn, or explore research code on GitHub.