Education
2014 - 2018 Imperial College London | MEng Computing (AI)
- Awarded the Governors' Prize for the top graduating MEng student in 2018 (87.5%, best overall in class).
- Other prizes: Formicary Software Engineering Prize, G-Research Prize (twice), Morgan Stanley IT Prize.
Work Experience
2023 - Present Kontext | Co-founder, CTO
- Founded an AI-native advertising platform that uses LLMs to serve contextually relevant ads inside GenAI apps; raised ~$10M seed and now lead a team of 7 engineers.
- Built the full serving stack on our own finetuned models running on vLLM on custom GPU infrastructure, delivering 20M+ hyperpersonalized ads/day and 2k+ RPS of DSP traffic.
- Designed the eval harness and ran post-training (SFT + DPO + GRPO) on open-weight base models to deliver brand-safe, on-policy ads even in NSFW and adversarial contexts.
2019 - 2023 Deepnote | Founding CTO
- Co-founded Deepnote, a collaborative data science notebook; architected the original platform (still in production today).
- Grew engineering from 2 to 20 engineers through Seed and Series A, building out processes as the company scaled.
- Scaled to 500k+ users across enterprise (Fortune 500), regulated industries (e.g. Norway's sovereign wealth fund), non-profits (e.g. Gates Foundation), and academia (e.g. MIT robotics).
- Owned SOC 2 Type II and HIPAA-compliant deployments to unblock regulated enterprise sales; built out security, compliance, and multi-cloud deployment infrastructure.
2018 - 2019 Two Sigma | Software engineer
- Implemented new execution scheduler significantly increasing model prediction speed (Java).
- Adapted internal build mechanisms to work with new Python APIs.
2017 Palantir | Software engineer (intern)
- Full stack development of data analytics tool for a customer (Java, Typescript, ElasticSearch)
- Implemented recommendation system based on the content similarity of analyzed documents
- Significantly improved search relevance and performance, improved data pipeline reliability
- Iterated with users on functionality and UX of multiple features during onsite visits
2016 Google | Site reliability engineer (intern)
- Created a system for analyzing per video resource usage in YouTube ContentID system
- Built an AB load-testing infrastructure using traffic replays for a new ContentID index
Publications
Matas, J., James, S., Davison, A.J. (2018). Sim-to-Real Reinforcement Learning for Deformable Object Manipulation. in Conference on Robot Learning (CoRL) 2018 - 500+ citations - arxiv.org/abs/1806.07851
Notable awards
- Forbes 30 Under 30 Slovakia
- O-1 Visa - US "extraordinary ability" classification
Notable projects
- Learning end-to-end robotic manipulation of deformable objects (Python, Tensorflow, Physics simulators)
- People’s choice award for best final year project - arxiv.org/abs/1806.07851
- Global Air Traffic Management For UAVs - in association Microsoft (C#, Python)
- Palantir Forward Group Project Prize for outstanding third year group project - news.microsoft.com
- Neural networks - character recognition and traveling salesman problem using recurrent neural network (Java)
- CPP prize for best project in category, team leader (5 imperial students)
- Improving Code Completion with Machine Learning (GPT2, LSTM)
- Proposed and led final year project for an Imperial student imperial.ac.uk (PDF)
- AREROS - autonomous rescue robotic system able to navigate in dangerous environments
- Intel ISEF 2012 – Finalist, World championship in robotics RoboCup jr. 2011 - 1st place
Skills
- Python
- TypeScript
- Java
- SQL
- LLM post-training (SFT, DPO, GRPO)
- Evals
- PyTorch
- vLLM
- ClickHouse
- Postgres
- Kubernetes
- GCP
- AWS
- Compliance
- Hiring
- Eng Management