We study how AI systems fail people, and build the evidence to fix them.

Software, benchmarks, and demonstrators we build and maintain. Everything here is public and free for research use.

• unbias-plusTool

  Bias detection and rewriting toolkit built on a fine-tuned Qwen3-4B. Classifies bias type and severity, locates biased spans, and returns a neutral rewrite in one structured output. Demonstrated to the Prime Minister of Canada in June 2026.

  PyPI Documentation Models
• FairSense-AgentiXProof of concept

  Agentic fairness and risk analysis platform: ReAct reasoning loops, dynamic tool selection, and self-critique over text, images, and datasets. FastAPI backend with streaming and a React front end.

  Project page
• FairSense-AITool

  Toolkit for bias auditing, emissions reporting, and explainability across large-scale AI systems, with responsible-AI compliance checks.

  Project page
• SONIC-O1Benchmark

  Human-verified benchmark for audio-video understanding: 231 videos, about 60 hours, 4,958 questions across 13 domains, with results reported by demographic group. Public leaderboard.

  Project page Dataset Leaderboard Paper
• HumaniBenchBenchmark

  Human-centred benchmark for large multimodal models organised around seven principles, including fairness, empathy, robustness, and multilingual ability. Presented at HAICON 2026.

  Paper
• VLDBenchBenchmark

  Benchmark for multimodal disinformation detection with regulatory alignment. Published in Information Fusion (2026).

  Project page Paper
• Data and Impact Accounting (DIA)Tool

  Tracks the cumulative carbon and water footprint of open-source models and their derivatives, with a public dashboard. ICML 2026 spotlight position paper; keynote at GREEN-AI, ECML-PKDD 2026.

  Dashboard Paper
• Educational visualisersProof of concept

  Interactive teaching tools accepted to the NeurIPS 2025 Educational Program: the climate cost of AI expressed in human-scale equivalents, and model immunisation, prompt-injection defence, and red-teaming explained through the vaccine analogy.

  From Data to CO₂ How AI vaccines work
• GreenBenchBenchmark

  Benchmark for the energy footprint of AI models across training and inference settings.

  Paper
• Agentic transparencyFramework

  Survey and framework for recording what an agent did and why, with the Minimal Explanation Packet as an audit-ready record at the end of each task.

  Project page Paper
• Responsible agentic reasoning (R²A²)Proof of concept

  Agents that carry fairness, privacy, and auditability checks through each reasoning step rather than only at the final output.

  Paper Code
• Dbias, NBias, and GUS-NetTool

  Earlier bias-detection releases: a configurable pipeline for news text (Dbias), token-level bias entities (NBias), and span-level generalisation, unfairness, and stereotype tagging (GUS-Net).

  Dbias on PyPI Dbias code GUS-Net models

Our work falls into four themes: fairness and bias in text and multimodal models; evaluation and governance of AI agents; synthetic media and deception; and the environmental cost of AI. A selection is below; work under anonymous review is not listed until it is accepted. Full lists are on Google Scholar under each author.

Agents: evaluation and governance

1. Evaluating and regulating agentic AI: a study of benchmarks, metrics, and regulation. Farooq, Raza et al., Emmanouilidis. Information Fusion, 2026.
2. TRiSM for agentic AI: a review of trust, risk, and security management in LLM-based agentic multi-agent systems. Raza, Sapkota, Karkee, Emmanouilidis. AI Open, 2026.
3. Transparency in agentic AI: a survey of interpretability, explainability, and governance. Raza, Radwan, Chaduvula, Alinoori, Emmanouilidis. Preprint, 2026.
4. From features to actions: explainability in traditional and agentic AI systems. Chaduvula, Ho, Kim, Narayanan, Alinoori, Garg, Ramachandram, Raza. Preprint, 2026.
5. Who is responsible? The data, models, users or regulations? A comprehensive survey on responsible generative AI for a sustainable future. Raza, Qureshi, Zahid et al. ACM Computing Surveys, 2025.
6. Developing safe and responsible large language models. Raza, Bashir et al. Machine Learning, 2025.

Fairness and bias

1. FairLens: benchmarking fairness in vision-language models for high-stakes decision-making. Raza et al. Preprint, 2026.
2. Bias in the picture: benchmarking VLMs with social-cue news images and LLM-as-judge assessment. Narayanan, Khazaie, Raza. NeurIPS 2025 workshop (Evaluating the Evolving LLM Lifecycle).
3. ViLBias: detecting and reasoning about bias in multimodal content. Raza et al. Preprint, 2025.
4. SONIC-O1: a real-world benchmark for evaluating multimodal LLMs on audio-video understanding. Radwan, Emmanouilidis, Tabassum, Pandya, Raza. 2026.
5. HumaniBench: a human-centric framework for large multimodal models evaluation. Raza et al. NeurIPS 2025 poster; under revision for ACM TIST.
6. Beyond content: how grammatical gender shapes visual representation in text-to-image models. Raza et al. EMNLP 2025 Findings.
7. LinguaMark: do multimodal models speak fairly? A benchmark-based evaluation. Raval, Narayanan, Khazaie, Raza. ASONAM 2025.
8. RESPECT: a framework for promoting inclusive and respectful conversations in online communications. Raza et al. Natural Language Processing Journal, 2025.
9. MBIAS: mitigating bias in large language models while retaining context. Raza, Raval, Chatrath. ACL 2024 WASSA workshop.
10. NBIAS: a natural language processing framework for bias identification in text. Raza et al. Expert Systems with Applications, 2023.

Synthetic media and deception

1. VLDBench: evaluating multimodal disinformation with regulatory alignment. Raza, Bashir, Emmanouilidis, Shah et al. Information Fusion, 2026.
2. F-DPO: reducing hallucinations in LLMs via factuality-aware preference learning. Raza et al. ACL 2026 Findings.
3. Just as humans need vaccines, so do models: model immunization to combat falsehoods. Raza et al. IJCNN / IEEE WCCI, 2026.
4. Fake news detection: comparative evaluation of BERT-like models and large language models with generative AI-annotated data. Raza, Paulen-Patterson, Ding. Knowledge and Information Systems, 2025.
5. FakeWatch: a framework for detecting fake news to ensure credible elections. Raza et al. Social Network Analysis and Mining, 2024.
6. Fake news detection based on news content and social contexts: a transformer-based approach. Raza, Ding. International Journal of Data Science and Analytics, 2022.

Environmental cost of AI

1. Sustainable open-source AI requires tracking the cumulative footprint of derivatives. Raza et al. ICML 2026, spotlight position paper.
2. Optimizing large language models: metrics, energy efficiency, and case study insights. Khan, Motie, Kocak, Raza. Preprint, 2025.
3. FairSense-AI: responsible AI meets sustainability. Raza, Chettiar, Yousefabadi, Khan, Lotif. Preprint, 2025.

Recommender systems and applied work

1. Review-based recommender systems: a survey of approaches, challenges and future perspectives. Hasan, Rahman, Ding, Huang, Raza. ACM Computing Surveys, 2026.
2. A comprehensive review of recommender systems: transitioning from theory to practice. Raza et al. Computer Science Review, 2026.
3. Progress in context-aware recommender systems: an overview. Raza, Ding. Computer Science Review, 2019.
4. Large-scale application of named entity recognition to biomedicine and epidemiology. Raza, Reji, Shajan, Bashir. PLOS Digital Health, 2022.
5. Clinical Application of Detecting COVID-19 Risks: A Natural Language Processing Approach. Bashir, Raza, Kocaman, Qamar. Viruses, 2022.
6. A jamming attack detection technique for opportunistic networks. Singh, Woungang, Dhurandher, Khalid. Internet of Things, 2022.
7. Reinforcement learning-based fuzzy geocast routing protocol for opportunistic networks. Khalid, Woungang, Dhurandher, Singh. Internet of Things, 2021.

In progress

• Reproducibility of bias-detection tools

  An audit of whether span-level bias detectors give the same verdict when the sampling seed, inference platform, or surrounding text changes.

• Harness-aware evaluation of agents

  A survey arguing that an agent's benchmark score belongs to the whole evaluation set-up (model, harness, environment, and evaluator), not the model alone.

• Deepfake forensics

  Audio-visual deepfake detection that generalises across generators: a survey, a trace-labelled forensic benchmark, and detection models that route on signal reliability.

• Does cheaper evaluation change the conclusion?

  Whether quantisation, batching, and benchmark reduction preserve conclusions about accuracy, fairness, and bias, and what each saves in energy and water.

Board of directors

• SB

  Dr. Syed Raza Bashir

  Founder, Chair of the Board

  Founder and Chair of the Board of SRLab.ai, with more than 25 years of experience across applied AI, software engineering, project and program management, higher education, research, and technology leadership.

  More

  Holds a PhD in Computer Science from Toronto Metropolitan University and is a PMP-certified professional with expertise in machine learning, responsible AI, software engineering, recommender systems, biomedical informatics, and fairness and bias in AI. Has led multidisciplinary technical projects, managed end-to-end research and software initiatives, and mentored students and engineering teams across academic, industry, and publishing environments, including as Scientific Managing Editor in Computer Science at Elsevier. Visiting faculty and capstone supervisor at Sheridan and Humber colleges. Research record includes publications in multimodal disinformation, responsible large language models, AI fairness, privacy, healthcare AI, and recommender systems. Provides strategic and governance leadership, helping shape the lab's research direction, partnerships, and mission to advance impactful, responsible, and applied AI research.

• SR

  Dr. Shaina Raza

  Chief Scientist

  Applied Machine Learning Scientist in Responsible AI at the Vector Institute and co-lead of Work Package 3 of the Horizon Europe AIXPERT project. Former CIHR Health System Impact Fellow working on equity in public-health AI. Responsible AI Leader of the Year (Women in AI, North America, 2025), one of The Peak's Emerging Leaders 2026, and on the Stanford/Elsevier top 2% list. Sits on CIHR and Killam peer-review committees. Principal investigator on the lab's research programme.

• KK

  Dr. Khurram Khalid

  Director and Scientist

  Researcher in opportunistic networks, secure routing, and energy-efficient protocols, with publications in Internet of Things (Elsevier) and IEEE venues. Leads the lab's work on secure and resilient systems.

• SF

  Shujaat Feroze

  Director, IT Infrastructure

  M.S., University of Wollongong. More than a decade of enterprise IT leadership at Telstra, Qantas, Saunders International, and Comcare. Responsible for the lab's cloud, security, and operational infrastructure.

Scientific advisory board

• RQ

  Dr. Rizwan Qureshi

  Scientific Advisor

  Senior Research Fellow at Massachusetts General Hospital, Harvard Medical School, where he works on foundation models for women's health and sex-specific disease. Previously at MD Anderson Cancer Center and the University of Central Florida.

  More

  Senior member of IEEE and DAAD AI Fellow; about 6,000 citations. Co-author of the lab's ACM Computing Surveys review of responsible generative AI.

• AV

  Prof. Athanasios Vasilakos

  Scientific Advisor

  Distinguished Professor at the Center for AI Research, University of Agder, Norway, and Dean of Class VI of the European Academy of Sciences and Arts.

  More

  Web of Science Highly Cited Researcher with more than 80,000 citations across AI, cybersecurity, and the Internet of Things. Advises on European partnerships and research direction.

Management and operations

• IL

  Imran Liaquat

  Advisor, Strategy and Operations

  Management consultant with more than two decades of international experience helping organisations develop and execute strategy, transform operations, and deliver complex business initiatives. A PMI Authorized Trainer, he has trained and mentored thousands of professionals across North America, Europe, and the Middle East.

  More

  His expertise spans strategy development and execution, the Balanced Scorecard, and KPI and OKR frameworks, together with portfolio, programme, and project management, risk and quality management, and organisational change. At SRLab he advises on organisational strategy, governance, operational effectiveness, and funding decisions, helping align the research portfolio with sustainable growth, strategic priorities, and partnerships.

Students and interns

Graduate students and research interns join the lab for defined projects, usually leading to a co-authored publication or a public release.

• MR

  Mahveen Raza

  Research Intern

  Works on AI sustainability, bias in generative models, and deepfake detection.

  More

  Co-author of four NeurIPS 2025 workshop papers (GenProCC and Muslims in ML), including studies of carbon literacy for generative AI and occupational stereotypes in text-to-image models, and lead author of the lab's NeurIPS 2025 educational materials, which she presented in San Diego. Currently working on the deepfake forensics programme. Portfolio

Institutions our members belong to or collaborate with. Listing here does not imply endorsement by the institution.