AI robustness & safety audit
Evaluate behavior under distribution shift, unknown inputs, adversarial manipulation, backdoors, and failure-prone operating conditions.
Available for selected consulting
Turn advanced AI research into dependable real-world systems.
I advise companies and public organizations on trustworthy AI, computer vision, anomaly detection, vision-language models, and robust machine learning—from independent technical review to focused proof-of-concept development.
Consulting is conducted in English. Local-language coordination can be arranged when needed.
Research-backedPeer-reviewed methods and grant-funded programs
InternationalExperience across six countries
IndependentEvidence-first technical assessment
FlexibleAdvisory, audit, workshop, or proof of concept
How I can help
Focused engagements for teams that need senior technical judgment before investing heavily in an AI direction.
Visual anomaly detection
Design approaches for defect detection and quality inspection when abnormal examples are rare, changing, or poorly labeled.
Computer vision & VLMs
Assess model choices, data strategy, evaluation design, and deployment risks for modern vision and vision-language systems.
Proof of concept
Translate a business problem into measurable objectives, a defensible baseline, experiments, and a go/no-go technical recommendation.
Technical due diligence
Independent review of AI claims, research proposals, model evaluations, technical roadmaps, and research-to-product feasibility.
Workshops & team advising
Practical sessions for engineering, research, product, and leadership teams on trustworthy AI and reliable evaluation.
Industry-relevant research portfolio
Selected research directions reframed around the operational problems they can help companies address. These are research foundations, not claims of client deployment.
CRANE: zero-shot anomaly detection
Context-guided prompt learning and attention refinement for recognizing anomalies without requiring a large set of labeled defects.
Potential use: manufacturing quality control, surface inspection, equipment and infrastructure monitoring.
FrameShield: robust video anomaly detection
Research on making video anomaly detection less vulnerable to adversarial or corrupted visual inputs.
Potential use: safety monitoring, video analytics, critical-site observation, and robust operational alerts.
Universal novelty detection
Adaptive contrastive learning for detecting previously unseen data across datasets and tasks.
Potential use: deployment monitoring, data-quality control, model fallback triggers, and unknown-category discovery.
International experience
Academic and research experience across distinct technical cultures helps me bridge research quality, practical constraints, and cross-border teams.
UzbekistanNew Uzbekistan University
Associate Professor in Tashkent, developing trustworthy AI research and local collaboration.
JapanOIST
Staff Scientist and Visiting Researcher; principal investigator of Japanese JSPS-funded AI research.
TürkiyeIstanbul Technical University
Visiting Faculty Member teaching deep learning and AI safety and security.
FinlandUniversity of Oulu / CMVS
Research experience in machine vision, anomaly detection, and reliable learning.
FranceResearch collaboration in Troyes
International research experience connecting machine learning and visual intelligence.
IranIPM and academic research
Faculty and research leadership experience in artificial intelligence.
A practical engagement
Start small, define evidence, and expand only when the results justify it.
Problem framing
Clarify the decision, risk, data, constraints, and success metric.
Technical assessment
Review data, models, evaluation gaps, and realistic solution options.
Focused validation
Run an audit, prototype, experiment plan, or proof of concept.
Decision & handoff
Deliver findings, limitations, recommendations, and a practical roadmap.
Have an AI problem where reliability matters?
Send a short description of your organization, the problem, available data, desired outcome, and timeline. I will reply if the project is a strong fit for my expertise.