Andrii RohovyiCo-Founder & CEO of Postdata AI
AI engineer with 11+ years in data science and ML, specialising in search, ranking, scoring and routing systems; PhD candidate at UNSW Sydney under Professor Toby Walsh.
Bio
Andrii is an AI engineer with more than eleven years in data science and machine learning. He specialises in search, ranking and scoring systems and has built them for telecom, government, maritime, logistics, finance, e-commerce and retail companies, including the NSW Government, the Austrian Economic Chamber, PayPal, BNP Paribas, Ralph Lauren, Macy’s, Kyivstar, Veon, Etisalat, OLX Group and Vodafone.
His research interest is search and pathfinding in transport. As a PhD candidate at UNSW Sydney, supervised by Professor Toby Walsh, he works on real-time public transit routing with unlimited transfers. A pruning method from this research, 457 lines of code including tests, was merged into OpenTripPlanner, the most widely used open-source transit routing engine, and made its targeted routing up to two times faster for every deployment. Transport agencies reported the gains in production; Skånetrafiken in Sweden saw average response times fall by at least 20%.
Before founding Postdata AI he was a lead data scientist at Grid Dynamics and a data science consultant at OLX Group, built maritime pathfinding at Ninety Percent of Everything, and started in data science at Kyivstar (Veon), DataRoot Labs and BNP Paribas. He holds a Master’s and a Bachelor’s degree in System Analysis from NTUU “KPI” (IASA), Kyiv.
Key projects
PayPal
Lead data scientist on the PayPal account, running a team of three data scientists, a Python developer and a DevOps engineer. Models that predict customer complaints and churn before they happen.
Vodafone
Data science consulting for the mobile operator.
BNP Paribas
Financial analysis, project profitability modelling and management reporting for UKRSIBBANK BNP Paribas Group, a bank with 300 branches, 1.9 million retail customers and 2,200 large corporate clients across Ukraine.
Ralph Lauren
Time-series forecasting of store traffic and demand (Prophet, SARIMA) to plan staffing and stock across Ralph Lauren’s retail network.
Macy’s
A search engine and ranking system for Macy’s built on LambdaRank and RankNet, lifting NDCG with product variability signals; visual search; a demand forecasting pipeline.
Etisalat
Data science consulting for one of the largest telecom groups in the Middle East.
OLX Group
New ranking models released in Kazakhstan, Ukraine, Uzbekistan and Romania with causal revenue analysis of each release; a multi-objective ranking system that lifted NDCG for visitors and promoted listings; personalised recommendations for Poland and hybrid image-and-text search for Portugal.
Kyivstar (Veon)
Credit scoring on telecom data, lookalike audiences with PU-learning, anti-fraud systems for a partner bank, churn and revenue forecasting, PII removal with NER, heatmaps for new outlets and A/B testing across 26 million subscribers of Ukraine’s largest operator.
NSW Government
Chatbots with safety guardrails for extracting metadata from documents, searching internal documentation and drafting briefing notes and statements of work from templates, plus an auditable legal document classification pipeline. Python, Azure, OpenAI and Azure AI Foundry.
Austrian Economic Chamber
Designed the search engine for the WKO website, cut p99 API latency from 8 s to 0.7 s, automated long-running data pipelines on GCP with Terraform and built an OpenAI-based content recommendation engine. A 3+ year engagement, still evolving.
Read the case →What people say
Scroll sideways →More
Other work14
- OpenTripPlanner: routing up to 2× faster
An early-pruning method for the core routing algorithm, 457 lines of code including tests, merged upstream and on by default. Skånetrafiken reported response times down by at least 20% in production.
Read the case → - HowToGet: a statewide journey planner for New South Wales
Nineteen Transport for NSW feeds, OpenTripPlanner and unlimited-transfer routing research in one live product that reaches places the official planner cannot.
Read the case → - Credit scoring for MD Finance and Mission Mobile
An end-to-end scoring system for the Romanian market with about 70% AUC, the best among the contractors evaluated; and an explainable LLM-based Financial Health Score from bank statements.
Read the case → - Ninety Percent of Everything, 2019–2020
Vessel routing from scratch with A*, HPA*, Theta* and Lazy Theta* on spherical geometry, with weather and bathymetry constraints; crew optimisation with linear programming; fuel prediction.
- The Kyiv Independent
A recommendation system with automated news abstracts and SEO extraction for the newsroom.
Read the case → - Eric AI
An educational avatar chatbot for the Hispanic community in the US: grounded RAG answers and real-time lip-synced video.
Read the case → - Chatbots with safety guardrails
An airport assistant in Thai and English, a multilingual educational platform in six languages, a personal assistant that resolves multi-hop reasoning, and a tattoo-generation creative tool.
- LLM-based application generator
Generates Node-RED and ToolJet applications with SQL from a prompt. Presented at Web Summit 2025; led a team of five developers.
- Delivery and vehicle routing
Real-time delivery planning with OR-Tools: vehicle routing, driver assignment and time windows.
Read the case → - Drone construction monitoring
3D reconstruction from drone data with OpenDroneMap and a VLM that compares imagery with the construction plan.
Read the case → - The APP Solutions, 2020–2022
Data science consultant: project estimation, MVP teams, and a YOLO system for clothes detection and recommendations from user photos.
- Social influencer platform
Forecasting likes and views for posts before publication.
- DataRoot Labs, 2017
Bookimed, a clinic-selection platform with 300,000 monthly users: user segmentation, related-page recommendations and key-question extraction from text.
- Open source and teaching
Reference implementations of the Timetable Nodes and unlimited-transfer routing papers on GitHub, and a lecture on shortest-path algorithms from Dijkstra to ANYA.
Open ↗
Blog posts5
- LLM-based credit scoring from bank statements →Postdata AI · 2026
- How we sped up public transit routing around the globe →Postdata AI · 2026
- Real-time public transit routing with unlimited transfers →Postdata AI · 2026
- Data Science: The Game Changer in Modern Finance →Postdata AI · 2026
- Demand sensing and forecasting: a unified solution for retail and manufacturing Grid Dynamics blog · 2024
In the press4
- UNSW researcher accelerates journey planning for millions of public transport users UNSW Newsroom
- Small code tweak leads to big journey mapper saving Spatial Source
- UNSW researcher accelerates journey planning for millions of public transport users The National Tribune
- UNSW researcher accelerates journey planning for millions of public transport users Medianet Newshub
Publications7
- Adapting Dijkstra for Buffers and Unlimited Transfers The IEEE International Conference on Intelligent Transportation Systems (ITSC), 2026
- Fast and Memory Efficient Multimodal Journey Planning with Delays The Symposium on Combinatorial Search (SoCS), 2026
- Early Pruning for Public Transport Routing The 16th World Conference on Transport Research (WCTR), 2026
- Multimodal Pathfinding with Personalized Travel Speed and Transfers of Unlimited Distance 2025 IEEE 37th International Conference on Tools with Artificial Intelligence (ICTAI), 2025
- Timetable Nodes for Public Transport Network arXiv preprint arXiv:2410.15715, 2024
- Fuzzy methods in the cognitive modelling task (Нечіткі методи в задачі когнітивного моделювання) International Scientific Journal, 2016
- Models for credit risk assessment (Моделі для оцінювання кредитного ризику) International Scientific Journal, 2016
Education3
- PhD candidate, Computer ScienceUNSW Sydney · 2023 – present
- Master’s, System AnalysisNTUU “KPI”, IASA, Kyiv · 2016 – 2018
- Bachelor’s, System AnalysisNTUU “KPI”, IASA, Kyiv · 2012 – 2016
Certifications22
- Claude Code in Action Anthropic · 2026
- Claude with the Anthropic API Anthropic · 2026
- Introduction to Model Context Protocol Anthropic · 2026
- Introduction to agent skills Anthropic · 2026
- Core Designer Certificate Dataiku
- AI For Everyone Coursera · 2019
- Introduction to Machine Learning Coursera · 2019
- Applied Problems of Data Analysis Coursera · 2019
- Drawing Conclusions from Data Coursera · 2019
- Finding Structure in Data Coursera · 2019
- Supervised Learning Coursera · 2019
- Deep Learning Specialization Coursera · 2018
- Sequence Models Coursera · 2018
- Convolutional Neural Networks Coursera · 2018
- Structuring Machine Learning Projects Coursera · 2018
- Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization Coursera · 2018
- Neural Networks and Deep Learning Coursera · 2018
- Big Data School 2.0Kyivstar · 2017
- Mathematics and Python for Data Analysis Coursera · 2017
- Data Science and Machine LearningDataRoot Labs · 2017
- Neural Networks for Machine Learning Coursera · 2017
- Machine Learning Coursera · 2016