01 / Retail
Shelf attention without lab hardware
Built a detection, tracking, and shelf-level gaze pipeline using standard cameras, pretrained face and gaze models, and visual embeddings for planogram heatmaps.
CTO / Data Scientist / Applied AI
Illya Tykhonov
I build computer vision and machine learning systems that survive real field conditions.
Full-cycle ML lead with a PhD background from MIPT. I work across data collection, annotation QA, model training, deployment, production support, and hiring - mostly in retail video analytics, gaze estimation, document automation, and mobile CV.
14 cm
mean shelf-level gaze localization error
92%
of laser-reference accuracy in timber measurement
40+
full ML and DL courses delivered
Profile
I am most useful where ML touches physical reality: cameras, noisy data, field setup, product constraints, and business-facing analytics. I can move from research notebook to deployed pipeline without losing the practical details.
Selected Work
Recent work distilled from the updated CV. The focus is on measurable systems, not generic AI claims.
02 / Mobile
AI eye-tracking for e-commerce UX
Led a consented mobile study pipeline with long-screenshot heatmaps and page-level attention analytics for product grids, filters, and creatives.
03 / Field CV
Timber measurement from phone photos
Released mobile CV pipelines for log-end detection, log counting, volume estimation, and timber type or grade classification on iOS and Android.
04 / Documents
OCR and NLP for enterprise workflows
Built table OCR and BERT-based document understanding workflows for procurement, supplier PDFs, NER, clustering, classification, and search.
Experience
Technical leadership plus hands-on delivery across production ML, computer vision, document AI, and education.
Aug 2022 - Present
Batumi / full-time
Batumi / full-time
Chief Technical Officer
Neuro Metrix
- Lead a full-cycle ML product stream in retail video analytics: study design, data collection, annotation QA, retraining, deployment, production support, and technical hiring.
- Productionized shopper attention pipelines and shelf heatmaps for planogram optimization with 14 cm mean localization error.
- Owned real-world field deployment: routers, cameras, on-site setup, and 3D-printed camera enclosures.
- Developed data-quality and identification models, including employee-vs.-non-employee classification with 98% F1.
Feb 2022 - Aug 2022
Istanbul / part-time
Istanbul / part-time
Chief Technical Officer
Neuro Wood
- Led end-to-end development of mobile computer vision for timber measurement automation from dataset creation to iOS and Android release.
- Built log-end detection, log counting, single-image volume estimation, and timber type or grade classification pipelines.
- Reduced field measurements from 15 minutes to 1 minute while retaining 92% of laser-reference accuracy.
Jul 2021 - Nov 2021
Moscow / part-time
Moscow / part-time
Data Scientist
Agroinvest Companies Group
- Developed a reinforcement-learning decision-support prototype for harvest equipment routing with PPO and Gym.
- Built scanned invoice table OCR and matched recognized items against internal purchasing databases.
- Improved BERT-based document workflows from 81% to 87% over TF-IDF plus logistic regression baselines.
Oct 2020 - Aug 2023
Moscow / full-time
Moscow / full-time
Lecturer
FPMI MIPT Innovation Lab
- Designed and delivered ML and deep learning curricula for open and enterprise programs.
- Delivered 40+ full-length courses to 2,000+ learners, including programs for Sber, Gazprombank, and VK.
- Led a 3-person team designing a standardized 4-year data analyst curriculum for technical schools and colleges.
Dec 2017 - Oct 2020
Moscow / full-time
Moscow / full-time
Data Scientist
Intellect LLC
- Developed forecasting models and a visualization interface for glaze production planning.
- Created Kohonen self-organizing map methods for unsupervised analysis of operational records.
- Built LDA-based document processing and workflow automation for thematic classification and routing.
Research & Patents
Selected publications and patents from computer vision, optimization, tracking, and adaptive systems.
- Tykhonov I.V. et al. High-Resolution Panoramic Video Synthesis and Object Detection on Panoramic Video. IRASET, 2024.
- Tykhonov I.V. et al. Two Ant Colony Optimization Approaches for Path Finding in a Sorting Center. IRASET, 2024.
- Tykhonov I.V., Zhdanov A.A. Adaptive Tracking Algorithm for Real-Time Applications. Neuroinformatics-2022, pp. 315-324.
- Tykhonov I.V., Zhdanov A.A. Knowledge Extraction Algorithm in Computer Vision Systems for Real-Time Applications. Neuroinformatics-2021.
- Kazakhstan patent No. 39401: determining gaze direction on a device screen while accounting for dynamic movement.
- Russia patent No. 2019616027: committee of self-organizing neural networks.
- Russia patent No. 2018663630: modular program of universal multilevel autonomous adaptive management of business processes.