Data Science Director at Agoda · Experimenting with AI and Robotics on the side · Successfully participating in ML competitions.
I'm Ilia Larchenko — Director of Data Science at Agoda in Bangkok, with 14 years across DS, ML and analytics. My teams ship high-load production systems for pricing and bidding optimization, customer-behavior and LTV prediction, agentic AI services and other projects. In my free time I work on robotics, ML and AI side projects.
In the past I was top-20 globally on Kaggle, more recently I won 1st places in robotics competitions: BEHAVIOR-1K (NeurIPS 2025) and LeHome (online round, ICRA 2026).
I post videos on YouTube and ship code on GitHub. I like to share my knowledge and explain complex technical problems in a clear and simple way.
Agoda · Bangkok
Leading multiple DS/ML teams across pricing, bidding, LTV attribution, experimentation, FinTech, bot detection, and LLM agents.
DOC+ · Moscow
Founded and led a 15-person DS/ML department at a health-tech startup. Built a pre-LLM medical chatbot symptom checker, electronic health record quality control and an automated doctor-dispatching system.
Consulting · startups · enterprise
Five years across analytics, strategy, finance, business development and operations — from consulting to startups and corporates.
Moscow Institute of Physics and Technology (MIPT)
Graduated with honors. Faculty of Control and Applied Mathematics.
1st place in the online round of the LeHome Challenge (ICRA 2026). The task was to train a bimanual robot in simulation to fold 4 different types of garments. I implemented the whole distributed RL pipeline from scratch, combining existing approaches with new ideas across model architecture, reward design, RL logic, and data collection.
YT
1st place in the BEHAVIOR Challenge at NeurIPS 2025 (from Stanford VL lab) — 50 long-horizon household tasks in the simulator. Our team of 3 trained a modified Pi0.5 VLA policy on demonstrations.
YT
25 medals in competitions, peak top-20 globally out of 150k+ users. Master tier in both Competitions and Notebooks.
3rd place out of ~900 teams in the first ARC Kaggle competition — the predecessor of today's ARC-AGI benchmark. Working in a team of 2, we each developed independent solutions and the final submission merged both into a single symbolic solver: a small DSL of grid transformations plus search. No ML, AI or LLMs.
YT
YT
YT
YT
YT
YT
YT
YT
YTDecoder-Only Transformer behavior-cloning policy. Sometimes 100× smaller than SOTA, no diffusion, no tokenization, no RL — very simple, but beats the prior best on PushT and ALOHA Insert.
YT
A Model Context Protocol server for the SO-ARM100 low-cost robot arm. Lets an LLM drive the arm directly through tool calls — no separate robotics policy required.
YT