Eric Hambro
Member Of Technical Staff at Anthropic
England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolEric Hambro is an NLP and probabilistic ML researcher-engineer with 11 years of experience building and evaluating large-scale language models, reinforcement learning agents, and Bayesian methods. Currently a Member of Technical Staff at Anthropic, he previously drove infrastructure, fine-tuning and offline RL research at Facebook AI and contributed to open-source benchmarks and libraries. His work spans practical production engineering—improving Gaussian process implementations in GPflow and adding rich observations and performance fixes to the NetHack Learning Environment—to research published in top conferences. Trained in experimental and theoretical physics (Cambridge) and machine learning (UCL), he combines rigorous mathematical grounding with hands-on systems optimization. A detail-oriented contributor, he has patched subtle GPflow dtype and SGPR bugs and migrated core methods to TensorFlow 2, showing a knack for catching hidden numerical issues.
11 years of coding experience
7 years of employment as a software developer
BA & MSci Experimental & Theoretical Physics, BA & MSci Experimental & Theoretical Physics at University of CambridgeUniversity College London
English, Portuguese, French, Russian
Github Skills (20)
python10
machine-learning10
datasets10
gaussian-processes10
tensorflow10
gpflow10
c-language9
bayesian-statistics9
terminal-application9
terminal-emulator9
performance-optimization9
bayesian9
cprogramming-language9
emulation9
deep-learning8
Programming languages (7)
Github contributions (5)
facebookresearch/nle
Nov 2020 - Dec 2022
The NetHack Learning Environment
Role in this project:
Back-end Developer
Back-end DeveloperContributions:152 reviews, 167 commits, 88 PRs in 2 years 1 month
Contributions summary:Eric primarily contributed to the enhancement of the NetHack Learning Environment by adding new observation features related to the dungeon environment. These additions included the implementation of text descriptions for objects on the map and terminal output, providing more detailed information about the game state. Furthermore, the user worked on performance improvements and benchmark tests, modifying code to reduce performance costs and optimizing screen description features. The user also fixed issues relating to the dataset and database.
multiagent-reinforcement-learninggame-aireinforcement-learningmahjongmachine-learning
GPflow/GPflow
Oct 2019 - May 2020
Gaussian processes in TensorFlow
Role in this project:ML Engineer
ML EngineerContributions:1 release, 22 commits, 38 PRs in 6 months
Contributions summary:Eric primarily focused on enhancing the Gaussian process functionality within the gpflow library. Their contributions include refactoring the `Module` class to align with TensorFlow 2 behavior and address a hidden bug in the SGPR model, involving diagonal likelihood calculations and the migration of the `compute_qu` method. They also ensured that tf_probability distributions used the correct dtype parameters and updated the MCMC notebook to leverage gpflow 2.0.
information-theorygpflowdeep-learningmachine-learningmarkov-chain-monte-carlo
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.