Research Engineer (AI / Systems / Reinforcement Learning)

GitHub • LinkedIn • dyllan@xlate.ai • dyllanmccreary@gmail.com (preferred)

Lead Research Engineer specializing in reinforcement learning, nanometer hardware design, simulation, and nearly all modalities in AI. Built and deployed superhuman RL agents at quilter.ai, contributed to Google Brain research, and attended NeurIPS — now founding the next generation of on-device and personalized AI at xlate.

Led development of quilter.ai’s reinforcement learning system, capable of designing circuit boards faster and more efficiently than human electrical engineers, combining:

• Reinforcement learning and game design
• Geometric constraint solving
• Large-scale simulation environments
• Novel training strategies outperforming internal state-of-the-art methods

Strong
Python • PyTorch • Reinforcement Learning • Deep Learning • Computer Vision

Experienced
C++ • Systems Engineering • Transformers • Computational Geometry

Exploring
Rust • Japanese • WebAssembly • UI/Design • iOS

Aug 2024 – Present
https://github.com/xlateai

Building next-generation systems for real-time, on-device AI training and interaction, focused on unifying learning, interfaces, and compute.

• Developed xlate.ai (https://xlate.ai), a language learning agent with per-token breakdowns and contextual understanding
• Built xos (https://github.com/xlateai/xos), an experimental application and AI framework
• Built Doom RL (https://github.com/verbiiyo/doom-rl)
• Created sensorlab (https://github.com/xlateai/sensorlab), a unified iOS sensor experimentation platform

May 2022 – Aug 2024
https://quilter.ai

• Trained and deployed a superhuman reinforcement learning agent for PCB component placement
• Built high-performance simulation environments generating millions of layouts per minute
• Engineered highly vectorized GPU systems for simulation and training

2022 – 2023
https://github.com/mindflowai/mindflow

• Co-developed an early LLM-based coding agent system
• Designed agent workflows for iterative code generation, execution, and feedback
• Explored early patterns in autonomous development systems prior to mainstream adoption
• Built abstractions similar to modern tool-using and self-improving agents

Dec 2021 – May 2022
https://gitcoin.co

• Designed adversarial simulations modeling strategic behavior in funding ecosystems
• Built systems to detect and mitigate fraudulent activity

Mar 2021 – Mar 2022
https://www.activeloop.ai/

• Architected systems for petabyte-scale dataset streaming
• Developed high-performance distributed training pipelines

Mar 2019 – Mar 2021

• Designed multi-agent RL for chip layout optimization
• Ran distributed training across 128+ GPUs

Nov 2018 – Mar 2019

• Conducted signal processing research
• Built early ML systems on real-world data

Jan 2017 – Feb 2019

• Built ML systems, game mods, and backend tools

• RigL Torch Contribution — https://github.com/verbiiyo/rigl-torch

  • Contributed to implementation and mathematical rigor
  • Referenced in Google Brain paper: https://arxiv.org/abs/1911.11134
  • Code cited in follow-up sparsity work: https://arxiv.org/html/2305.02299v4

• Multi-Agent Systems (PhD Collaboration) — https://github.com/verbiiyo/ml-arena

  • Contributed to research on adversarial multi-agent environments and dynamic programming
  • Work informed subsequent research directions (unpublished collaboration)

• MineRL BASALT Competition — https://github.com/verbiiyo/basalt_2022_competition_nollied

  • Participant in BASALT challenge
  • Invited and sponsored to attend NeurIPS via MineRL (the day ChatGPT launched!)

• Neuroevolution (Flappy Bird) — https://github.com/verbiiyo/neuro-evolution-flappy-bird

  • Evolutionary algorithm–based policy learning

• Reinforcement Learning Foundations — https://github.com/verbiiyo/ReinforcementLearning

  • GridWorld (TD learning)
  • Multi-Armed Bandit
  • TicTacToe (value-based methods)

2018+

• Machine Learning foundations (regression, optimization)
• Deep Learning & Computer Vision
• Reinforcement Learning fundamentals

2024 - 2024

• Formal study following 1 year of self-directed learning

2017 – 2018

• Founded and led Computer Science Club
• Student government for opening a new club
• Worked with the computer science department professors to allocate classroom time for us to meet
• Wrote video games collaboratively with our members in Lua and offered general guidance and support https://github.com/sierra-college-cs-club

2013 – 2017

• Completed AP Computer Science coursework with an A (college credit)
• Final project: built an Asteroids game in C++ (https://github.com/verbiiyo/Asteroids)
  • Received special permission to use C++ instead of the required Snap language
  • Went beyond course requirements with a full game implementation and performance-focused design

Wrote my first program in 2011 as a middle school student: a security system and factory management OS for the Minecraft modpack Tekkit using Lua (ComputerCraft). Iterated on this across multiple worlds, developing early intuition for automation, systems design, and stateful environments.

From 2014 through high school, focused heavily on C++, Java, Unreal Engine, and web systems, building games and tools (e.g., https://github.com/verbiiyo/Asteroids), which led to an interest in performance engineering and low-level optimization.

In 2018, began studying AI formally with Udemy (coursework). Progressed from:

• Linear and logistic regression
• Multivariate optimization and cost function derivation
• Deep learning and convolutional networks
• Reinforcement learning (value iteration, MCTS, TD methods)

Built early RL systems including:

• TicTacToe solvers (value iteration, MCTS, etc.)
• Multi-armed bandits and GridWorld control systems
• Transition into deep RL and ultimately multi-agent reinforcement learning

This trajectory led to specialization in RL systems, GPU-accelerated training, and evolutionary methods, forming the foundation of current work.