Profile
AI/ML executive with two decades of experience leading research and engineering at the frontier — from foundational RL research at DeepMind to production AI infrastructure at Meta, the Ellison Institute of Technology (Oxford), and the Tony Blair Institute. Repeat founder and founding-team operator, currently building an autonomous coding agent for the enterprise at seed stage. Track record of recruiting and scaling world-class technical organisations, setting scientific direction, and shipping novel AI products into production.
Experience
Pavo AI · Director of ML
Jan 2026 – Present
Co-leading technical strategy at a seed-stage startup building an autonomous, life-long learning coding agent for the enterprise.
- Standing up the research and engineering organisations from zero.
- Owning the technical and scientific roadmap for the core challenges in life-long learning.
Tony Blair Institute for Global Change · Head of ML, AI Incubator (Founding Team)
Mar 2025 – Nov 2025
Founding member of the leadership team that launched a new AI Incubator to deliver AI products into government partners worldwide.
Organisation leadership
- Set the AI product and ML infrastructure strategy for the Incubator, enabling rapid prototyping and deployment of AI applications across multiple government partners.
- Recruited and led an interdisciplinary team across software and cloud engineering.
- Built a portfolio of agentic AI prototypes spanning the Incubator’s initial offering.
Technical leadership
- Architected applications leveraging large language and deep learning models for globally distributed government use cases.
- Developed reinforcement-learning–based world-action models that capture relationships between real-world systems based on legislation, policy, and econometric data — modelling officials’ actions and simulating short- and long-term effects.
- Led research on improving accuracy and grounding of model predictions.
Ellison Institute of Technology, Oxford · Head of MLOps
Sep 2024 – Mar 2025
Founding member of the AI & Data team — a horizontal function enabling the entire institute’s AI and ML capabilities.
- Defined the ML infrastructure strategy supporting innovation and productionalisation across the institute’s four domains: medical sciences, climate, agriculture & sustainability, and government efficiency.
- Recruited and led an engineering team across software and cloud systems.
- Owned the strategic partnership with Oracle.
Leapfrog Studios Ltd · VP Applied Research, Co-Founder
Jul 2023 – Sep 2024
Co-founded a studio applying machine learning to invent new ways video games engage players. Historically, innovation in games has tracked hardware cycles — increased compute (PlayStation, Xbox) or new modes of interaction (mobile touch, Kinect, Wii, VR). Leapfrog was the thesis that, for the first time, algorithmic innovation alone now offers a comparable opening for new genres on PC and console.
- Recruited and led a team of research scientists and engineers.
- Set the scientific and product direction.
Meta · Staff Software Engineer
Mar 2022 – Jul 2023
Online experimentation infrastructure for Ads Ranking — Uber-Tech Lead for the area responsible for the correctness and accuracy of experiment results, with 20 engineers across six projects spanning pure statistical research to full-stack engineering. Operated against compounding headwinds: p80 of true effects below the platform’s detectability threshold, design that facilitated p-hacking, and complex feedback loops creating confounders. Set technical direction, ran roadmapping, coached engineers, and introduced large innovative bets to move the platform onto a new S-curve. (HHVM, Python, Thrift, Presto, EntFramework)
ML model delivery support — Single point of contact for a team R&Ding key improvements to the ML models powering Meta’s ranking system. Translated their needs into the org’s technical roadmap and assembled ad-hoc task teams to unblock delivery. Awarded the highest performance rating (“Greatly Exceeds Expectations”) in the first year for enabling the partner team to ship nearly 50% more innovation than originally planned.
DeepMind · Senior Software Engineer
Jun 2017 – Feb 2022
Model-based RL infrastructure — Tech lead of a project building high-performance, distributed infrastructure for training model-based RL agents in fully observable, continuous environments with high-dimensional observations. Defined technical direction, coordinated with other ML ecosystem leads, and developed alongside research scientists and engineers. (Python, C++, JAX, Cloud)
Rapid adoption of third-party RL technology — Owner and architect of infrastructure that significantly reduced the engineering effort to adopt third-party technology for generating RL training data. The largest project I led at DeepMind, coordinating across several teams; downstream projects launched with one engineer in hours. (C++, Python, Cloud, Docker)
Large-scale dataset for supervised learning — Tech lead and primary developer of infrastructure for collecting data at scale from hundreds of thousands of instances of a highly specialised scientific simulator. Adapted code designed for different distributed architectures to perform optimally on Google Cloud. (Python, C++, Docker, Cloud)
Robot perception task suite — Sole engineer in a research team studying alternative perception models for controlling robotic grippers in complex manipulation tasks. Built a suite of 3D environments and tasks for agent training. (C++, MuJoCo, Unity 3D, Cloud)
Causal reasoning task suite prototype — Lead developer in a multidisciplinary team building a prototype task suite for studying causal models. Defined the abstractions that allowed four engineers to focus on rapid prototyping. (C#, Unity 3D)
Using Unity to Help Solve Intelligence — Contributor on the team that adopted Unity 3D for AGI research; built the interfaces and protocols for exchanging data between agents and Unity environments. (C#, Unity 3D, Python)
DeepMind Control Suite — Owned the suite visualiser: implemented desktop and web (Colab) versions, open-sourced the project, and supported the user community. (Python, MuJoCo, Colab)
Botperfect · Owner, Software Engineer
Jan 2016 – Jun 2017
Kanban tasks manager for accountancy offices (own product) — An assistant for managing accountancy documentation fused with a comprehensive task tracker. Ran market research, built the MVP, deployed with first customers, and coordinated third-party contractors. (C#, JavaScript, HTML, ASP.NET, Bootstrap)
Hatred (commissioned by Destructive Creations) — Prototype of a twin-stick shooter focused on character animation and environment destruction. (C++, UE4, PhysX, 3DS Max)
Virtual battlefield simulation (commissioned by Bagira Systems) — Improved performance of animation and physics systems to simultaneously simulate >5,000 visible units. (C++, Havok Physics/Animation/Behavior)
PDF documents optimiser (commissioned by Digital Recycle Ltd) — A compression library for PDFs that preserved original quality. (C++)
Crytek Frankfurt · Senior Animation Programmer
Feb 2016 – Mar 2016
Built the locomotion system for the dinosaur characters in the VR title Robinson the Journey. The challenge: bipedal creatures with long tails moved in a snake-like way, so I combined standard bipedal techniques with novel approaches to capture that sleek motion. (CryEngine, C++)
Havok · Senior Software Engineer
Nov 2011 – Dec 2015
Built a deployable library of characters and vehicles for simulations: blend trees for locomotion, aiming, and on-demand poses; integration with Havok AI for navigation; client/server networked vehicles; content-authoring pipelines for 3DS Max and Maya; integration with Havok’s vForge engine. Designed a 2D parametric motion controller blending multiple clips against velocity/turn-angle inputs and a velocity-domain animation synchroniser preventing output-velocity distortion in blended locomotion. Implemented ragdoll controllers for the new physics library. Worked closely with animators and 3D artists. (C++, C#, LUA, Havok Behavior/Physics/Animation, Oculus Rift DK2)
Based in the Dublin office for the first year, then remote from Poland.
CD Projekt RED · Senior Gameplay/AI Programmer
Dec 2009 – Oct 2011
The Witcher 2 (Xbox 360) — AI subsystem optimisations for Xbox 360. Supervised work on gameplay features for the platform. (C++, wxWidgets, DirectX 9.0, in-house RedScript)
The Witcher 2 (PC) — character navigation and locomotion — Integrated PathEngine navigation with the in-house animation system. Implemented local steering / obstacle avoidance, kinematic constraints between navigation and animation, off-navmesh traversal, and docking to obstacles. (C++, PathEngine, Havok Animation/Physics, RedEngine)
The Witcher 2 (PC) — non-linear quest authoring system — A visual block language (similar to Unreal’s Blueprint) that let writers and designers transform gameplay ideas into shipped gameplay. The system orchestrated every other gameplay subsystem in the engine. End-to-end ownership: core system, editors and debuggers, integration, and ongoing user mentorship. (C++, wxWidgets, RedEngine)
The Witcher 2 (PC) — character animation, research, and engine — Animated all in-game characters with artists and designers. Researched formation movement, crowd simulation, and sensory systems. Worked on core RedEngine systems: a saves system integrated with a real-time multithreaded scripting language, world streaming, and performance optimisations.
Coversion · Owner, Software Engineer
Nov 2008 – Nov 2009
Multi-platform game engine for PC and Nintendo Wii (outsourced to Bloober Team) — Designed and developed input, basic animation controller, sound, serialisation, scene-graph management, and an FSM-based AI subsystem. (C++, Codewarrior)
Navigation system for Egypt: Engineering an Empire (Nintendo DS) (outsourced to Bloober Team, released by Slitherine) — Designed and developed a game editor for a turn-based strategy game. (C++, Qt)
Sabre Holdings Polska · Software Engineer
Nov 2006 – Oct 2008
Laid out the architecture for the next-generation data-mining application. Mentored teams on TDD and OOAD techniques. (C++, CppUnit)
Motorola Polska Electronic · Software Engineer
Mar 2005 – Oct 2006
Designed and developed a packet data router for a Tetra-standard telecommunication system. (C++, Linux, Bash, Tsh)
Publications
- Patent: Simulating Industrial Facilities for Control — Application no. PCT/EP2023/067148.
- Semi-analytical Industrial Cooling System Model for Reinforcement Learning
- Fast transport simulations with higher-fidelity surrogate models for ITER
- Using Unity to Help Solve Intelligence
- dm_control: Software and Tasks for Continuous Control
- Evaluating model-based planning and planner amortization for continuous control
- Learning Dynamics Models for Model Predictive Agents
- Augmenting learning using symmetry in a biologically-inspired domain
- “Unit tests vs. functional tests — round 101” — Gamasutra blog
- “Virtual mess — destructors in C++” — Gamasutra blog
- Articles on animation programming — ptrochim.wordpress.com
Education
- 1999–2005 — MSc, Electronics and Information Technology, Technical University of Lublin. Master thesis: Adaptive regulation of randomly disrupted plants.
Communication
- Programming languages: C++, Python
- Natural languages: Polish (native), English, Spanish, German, French
- Universal languages: Linear algebra, Calculus, Probability Theory