Taking AI Research from Lab to Market
GO! GLOBAL Türkiye Summit 2026, Istanbul. How research becomes a commercial opportunity: finding useful applications, validating demand and taking the first steps toward a company.
AI researcher, engineer and founder · Oxford and Istanbul
Taking AI research from the lab to the market, responsibly.
Dr Omer Gunes is the Founder and CEO of Oxtractor, an Oxford University Innovation portfolio company building multi-agent retrieval and conversational AI systems, and a co-founder of TradeComplize, an AI-native trade finance compliance platform backed by the Istanbul company builder VLABS34. He holds a DPhil (the Oxford PhD) in Computer Science from the University of Oxford, where he has worked for fifteen years, and teaches and supervises at UCL and Koç University.
15 yearsat Oxford Computer Science
45+MSc and PhD supervisions and co-supervisions at Oxford and UCL
2AI companies founded and co-founded
3universities: Oxford, UCL, Koç
On stage
This week in Istanbul and Bodrum
Speaking to founders and investors at GO! GLOBAL Türkiye, and to teachers at the International Teacher Summit organised by Endless Abroad.
Roundtable host
Frontier AI Research: What's Next and How Should Türkiye Compete?
GO! GLOBAL Türkiye Summit 2026, Istanbul. With panelists from Harvard and Apple, moderated by IBM.
When Intelligence Becomes Abundant
International Teacher Summit, organised by Endless Abroad. Redesigning the school for the age of reasoning machines: what AI can now generate, reason, do and model, and what school is for when intelligence is cheap.
Approach
One path, four stages
Every role I hold sits somewhere on the same path: a research question becomes a working system, the system becomes a product, and the product is deployed in a way people can trust.
STAGE 1
Research
Questions worth answering, pursued at Oxford, UCL and Koç with students and colleagues.
Research and teachingSTAGE 2
Engineering
Turning methods into systems that run on real documents, real users and real hardware.
EngineeringSTAGE 3
Ventures
Companies and partnerships that carry the work to organisations that need it.
Ventures and advisorySTAGE 4
Responsible deployment
Sovereign data, measured performance and human accountability, built in from the start.
Responsible AIResearch and teaching
Where the work starts
My research began in web-scale information extraction at Oxford and now centres on multi-agent retrieval-augmented generation, conversational AI, the evaluation methods that make these systems dependable, and responsible computing.
Current affiliations
- Senior Research FellowKUIS AI Centre, Koç University2025 to now
- Associate MemberResponsible Technology Institute, Department of Computer Science, University of Oxford2024 to now
- AI AdviserQuill Project, Pembroke College, Oxford
- MSc dissertation supervisorDepartment of Computer Science, UCL2020 to now
- Course Coordinator, AI Literacy: Foundations and ApplicationsKoç University, scaling to around 1,000 students a year
Education
- DPhil in Computer ScienceUniversity of Oxford. Supervised by Professor Georg Gottlob within the ERC-funded DIADEM project2010–2017
- MSc in Computer ScienceNew York University2009–2010
- BEng in Computer EngineeringGalatasaray University, with an Erasmus exchange at Université de Lille
Research themes
Supervision
I have supervised or co-supervised more than 45 MSc and PhD projects at Oxford and UCL. The current UCL cohort works on skill retrieval for LLM agents, multimodal RAG for enterprise documents, evaluation metrics for RAG systems, cross-lingual toxicity detection, continual learning, federated fine-tuning and tool-augmented agents for biomedical research. Several of these projects are now being prepared for publication.
Selected publications
- Investigating Human Trafficking Recruitment Online: A Study of Fraudulent Job Offers on Social Media PlatformsT. J. Moyo, O. Gunes, M. Jirotka · Proceedings of the ACM on Human-Computer Interaction, 2025
- A Responsible Research and Innovation Approach to Assessing the Implications of Anti-Trafficking TechnologiesT. Moyo, W. Albayaydh, H. Webb, O. Gunes, M. Jirotka · ACM Journal on Responsible Computing, 2025
- Towards Explainable and Trustworthy Collaborative Robots through Embodied Question AnsweringL. Kunze, O. Gunes, D. Hillier, et al. · 2022
- Ask George Washington on Negotiated Texts with Quill Q&AO. Gunes, N. Cole, S. Bates, J. Tupp-Mugglestone · IEEE ICDM Workshops, 2018
- Structured Aspect ExtractionO. Gunes, T. Furche, G. Orsi · COLING, 2016
- DIADEM: Domain-Centric, Intelligent, Automated Data Extraction MethodologyT. Furche, G. Gottlob, G. Grasso, O. Gunes, et al. · WWW Companion, 2012
Engineering
How it gets built
I still write and review the systems my teams ship. The common thread is multi-agent architecture: specialised agents for routing, retrieval, reasoning and compliance, with an independent judge checking the answer before anyone sees it.
J6 multi-agent RAG
A retrieval system in which a router sends each question to the right retrieval strategy, a reasoning agent assembles the answer, a compliance agent checks it against policy, and a separate judge model scores it. Presented as a technical deep dive at the Mindstone Cambridge AI Meetup.
LangGraph · ChromaDB · Elasticsearch BM25 · Neo4j · vision-language models for documents · LLM-as-judge
Oxtractor Nucleus
An industry-agnostic, model-agnostic platform on which Oxtractor's products are built. Sector knowledge arrives as domain packs, so the same core can serve healthcare, education, law or finance.
Foundation-model agnostic · domain packs · on-premise deployment
Deep Conversational AI
A conversational AI and AI tutoring platform for learning and research, co-designed with clinicians and educators, with a current focus on medicine and scientific training.
Multi-hop question answering · tutoring dialogue · evaluation pipelines
AI Scientist
A research programme with my UCL MSc students on AI agents that support scientific work, organised around three tests a laboratory should see passed before relying on one. Reading: can the agent find and use the right literature? Trusting: can its claims and reasoning be checked? Sharing: can institutions improve a model together without pooling their data?
Tool-augmented research agents · retrieval configuration · claim-level auditing · federated fine-tuning
Ventures and advisory
Carrying research to market
I build companies around problems where retrieval, reasoning and compliance have to work together, and I advise organisations that are making the same move.
Oxtractor
Founder and CEO
An Oxford University Innovation portfolio company and R&D studio for multi-agent retrieval and conversational AI. Oxtractor builds the Nucleus platform and Deep Conversational AI, works with university partners across Oxford, UCL and King's College London, and focuses on healthcare, education and other regulated domains.
oxtractor.comTradeComplize
Co-founder
An AI-native RegTech platform that helps banks review letter of credit document sets faster, checking them against ICC rules and UCP 600. Backed by VLABS34.
tradecomplize.comVLABS34
Board member
An Istanbul company builder that creates and funds AI, robotics and deep tech ventures in real-economy industries.
v-labs.aiKloia
Chief AI Adviser
Advising Kloia, an AWS Premier Tier Services Partner and cloud, DevOps and AI consultancy, on AI strategy and delivery for enterprise clients.
kloia.comAIPA
Member, education and healthcare commissions
An AI policy think tank in Ankara working on how Türkiye adopts AI in schools, universities and health services.
aipaturkey.orgWorking on an AI programme?
I take on a small number of R&D partnerships, AI transformation programmes and research consortia each year.
Responsible AI
Built to be trusted
Responsibility is a design decision, made early. These four principles shape the systems I build, the students I supervise and the policy work I contribute to.
Data stays where it belongs
Sensitive institutions should not have to send their data to someone else's cloud to use AI. Our systems run on premise or in sovereign infrastructure by default.
Oxtractor deployments are on premise with full data sovereignty.
Measure before you scale
A deployment is ready when its error rate is known, not when its demo looks good. Evaluation and hallucination control come before rollout.
Judge agents in J6 and supervised research on RAG evaluation metrics.
People stay accountable
Agents can draft, retrieve and check, but a named person owns the decision. Compliance checks and audit trails are part of the architecture.
Compliance agents in J6 and TradeComplize.
AI literacy is a public good
Societies adopt AI well when people understand what it can and cannot do. That starts in the classroom and continues in policy.
AI Literacy at Koç University, AIPA commissions, and research with Oxford's Responsible Technology Institute on online human-trafficking recruitment and anti-trafficking technologies.
Speaking
Talks and speaker kit
Keynotes, panels and roundtables for founders, investors, universities and public institutions. Everything an organiser needs is below.
On Turing's 1950 paper, Yıldız Technical University · AI in Education Summit, Istanbul · Koç University, Istanbul
Signature topics
- From lab to marketHow AI research becomes a product, a company and a deployment people trust.
- Multi-agent AI in regulated industriesRetrieval, reasoning and compliance in healthcare, finance and law.
- The AI scientistAgents that read, reason and run experiments alongside researchers.
- When intelligence becomes abundantRedesigning schools and universities for the age of reasoning machines.
- AI literacy for everyoneWhat students, professionals and policymakers need to know now.
- Türkiye in the frontier AI raceTalent, research links and the path to globally competitive companies.
Recent talks
- When Intelligence Becomes AbundantInternational Teacher Summit, Endless Abroad, Bodrum. KeynoteOct 2026
- Taking AI Research from Lab to MarketGO! GLOBAL Türkiye Summit, IstanbulOct 2026
- Frontier AI Research and TürkiyeGO! GLOBAL Türkiye Summit, IstanbulOct 2026
- The AI ScientistAziz Sancar Institute of Experimental Medicine, Istanbul UniversityOct 2026
- AI in Education SummitYıldız Technical University, IstanbulApr 2026
- Inside the J6 multi-agent RAG architectureMindstone Cambridge AI Meetup2026
Bios
Dr Omer Gunes is an AI researcher, engineer and founder working between Oxford and Istanbul. He is Founder and CEO of Oxtractor, an Oxford University Innovation portfolio company building multi-agent retrieval and conversational AI systems, and co-founder of TradeComplize, backed by VLABS34. He holds a DPhil (the Oxford PhD) in Computer Science from the University of Oxford and has supervised or co-supervised more than 45 MSc and PhD students.
Dr Omer Gunes is an AI researcher, engineer and entrepreneur with a fifteen-year affiliation with the Department of Computer Science at the University of Oxford, where he completed his DPhil (the Oxford PhD) under Professor Georg Gottlob on the ERC-funded DIADEM project. He is Founder and CEO of Oxtractor, an Oxford University Innovation portfolio company developing multi-agent retrieval-augmented generation and conversational AI through its Deep Conversational AI platform, and co-founder of TradeComplize, an AI-native trade finance compliance platform backed by VLABS34. He is a Senior Research Fellow at Koç University's KUIS AI Centre, an associate member of Oxford's Responsible Technology Institute, AI Adviser to the Quill Project at Pembroke College, and supervises MSc dissertations at UCL. He also serves on the board of VLABS34, as Chief AI Adviser to Kloia, and on AIPA's education and healthcare commissions. He studied at Galatasaray University, Université de Lille and New York University.
Photos
High-resolution images for programmes and press. Credit is not required.
Oxford, colour
Portrait, mono
Close-up, mono
Full length, colour
Contact
Speaking, partnerships, research and press
Write to me directly. I read every message and reply to most within a few days.
omer@oxtractor.com