I work with teams and organisations that need institutions read correctly and analytical output they can defend. If you face a government or international organisation you have to engage, owe a funder a measurement plan, produce analysis someone has to stand behind, are deciding where AI belongs in your operations and how to answer for it, or need to show what a programme or a policy actually changed when randomising is impossible, this page is for you.
For companies and institutions whose next constraint is not technical but institutional: a regulator to persuade, a ministry to read correctly, an international organisation to engage. Eight years as a career diplomat, from the WTO and UNCTAD floor in Geneva to the G20 file in Ankara, taught me how these institutions actually decide. I map the stakeholders who matter, watch the regulatory pipeline before it lands, and draft the briefs and positions principals actually use. You get external relations strategy from someone who sat on the institutional side of the table.
For organisations writing proposals or running programmes where a funder asks: how will you know it worked? I take the measurement side off your hands: outcomes defined so they can actually be detected, survey instruments, monitoring design, and the analysis plan that makes the evidence requirement satisfiable. You get the methods section a reviewer stops arguing with. Current work in this line is the measurement angle of the Values for Cohesion project at the University of Birmingham, funded by the UK Ministry of Housing, Communities and Local Government.
For teams that ship AI-assisted analysis someone senior has to sign. I build the review layer that makes that signature defensible: a pipeline where no model audits its own family's output, findings that cite their sources, and a written record of what was checked. You get a working review system on your corpus, not a slide deck. Built once as Project Shimmer; adapted to your domain.
For teams that need to know whether a model, a method or a published claim actually performs before money or reputation goes on it. I design the test: benchmarks with ground truth built by construction, de-primed setups that measure judgment rather than recall, and causal-integrity audits that ask whether a result survives the robustness checks its authors did not report. You get the evaluation design, the run, and a written verdict you can act on.
For organisations that must answer for their AI: to a regulator, a board, or the public. I work both sides of that divide, building multi-agent systems and writing on their governance. Readiness against the EU AI Act and the regimes following it, internal governance frameworks that engineers can actually implement, and position papers that hold up in front of lawyers and technical teams at the same time. You get governance that matches how the systems really behave, not a policy binder.
For organisations that know AI belongs in their work but not where, or how far to trust it. I map your document flows and decision points, identify where agentic systems and machine learning genuinely pay off and where they add risk, and design the governance layer: what gets automated, what stays human, and how output is verified before anyone acts on it. You get an adoption plan written by someone who builds these systems, not someone selling seats.
For policy units, research teams and social science projects that have to say whether something worked when randomising is impossible: did the programme change behaviour, did the measure move the sector it applied to, which groups responded and which only appear to. I bring the toolkit built for exactly these settings, difference-in-differences, instrumental variables, regression discontinuity, and double machine learning for heterogeneous effects, together with the diagnostics that decide whether the estimate survives. You get a defensible effect estimate from data you already have, and a written account of what would have to be true for it to be wrong.
Based in İzmir and working remotely by default, in English and Turkish, invoicing as an independent contractor. Hybrid and on-site arrangements are open to discussion where the work calls for presence. Scoped engagements preferred: a defined corpus, a defined question, a defined deliverable. Longer arrangements begin the same way.
Tell me what you are trying to verify, evaluate, measure or navigate, and what is at stake. I reply within two working days with a yes, a no, or the name of someone better placed.
If you prefer plain email, write to me directly.