Skip to content
View amcheste's full-sized avatar
9 followers · 35 following

Block or report amcheste

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
amcheste/README.md

Alan Chester · @amcheste — (ε, δ) — For any ε > 0, there exists a δ > 0.

For any standard, there exists a step that lands inside it. The work is finding it.

brand research writing

I work at the intersection of cloud infrastructure, agentic AI systems, reinforcement learning, and multi-agent trust. This profile is where the technical and research work lives.

What I build

Reinforcement learning

  • pokemon-red-ai · Reinforcement learning research on observation representations in Pokémon Red. Pixel, symbolic, and hybrid conditions under capacity-matched encoders. RecurrentPPO via SB3-Contrib.

Agentic systems

  • claude-teams-operator · Kubernetes operator that runs Claude Code agent teams as distributed pods.
  • ea-agent · AI personal executive assistant built around Obsidian.

Research tooling

  • overleaf-mcp · MCP server for editing Overleaf LaTeX projects from Claude. Published to PyPI. Single-user by design, auditable, built for academic researchers.
  • paper-skills · Claude Code skills for academic paper triage and Obsidian integration.

Applied experiments

Identity and tooling

  • alanchester-brand · Personal brand system, expressed as code. Tokens, components, the equation as identity.
  • engineering-handbook · Personal engineering handbook. Philosophies, workflows, and tooling for how I build software. Versioned with semver.
  • mac-dev-setup · One command from zero to productive on macOS.

What I publish

Research and writing on reinforcement learning, agentic systems, multi-agent trust, and team architectures for the agentic era. Currently drafting a paper for ARLET 2026 (NeurIPS workshop).

The question: in long-horizon reinforcement learning, do symbolic observations beat pixel observations once encoder capacity is properly controlled? Pokémon Red is the empirical environment.

The current questions

On reinforcement learning. How does observation representation affect sample efficiency in long-horizon reinforcement learning? Pixel versus symbolic versus hybrid observations under capacity-matched encoders are the conditions under study.

On team architecture. How does the unit of work change when teams include autonomous agents, and what structures support variable-elasticity teams?

On multi-agent security. How does trust topology affect security in multi-agent LLM systems? Centralized orchestrator versus peer-authenticated trust.

Background

About fifteen years across kernel engineering, Kubernetes platform development, cloud security, and product leadership. Currently leading a team of product managers at Oracle Cloud Infrastructure.

Dual B.S. in Computer Science and Applied Mathematics, NC State.

MBA candidate, NC State Jenkins Graduate School of Management.

Find me

δ ε

Pinned Loading

  1. alanchester-brand alanchester-brand Public

    Personal brand system. ∀ ε > 0, ∃ δ > 0.

    JavaScript

  2. pokemon-red-ai pokemon-red-ai Public

    Reinforcement learning toolkit for training agents to play Pokémon Red. Built on PyBoy, Stable-Baselines3, and Gymnasium. Includes pixel/symbolic/hybrid observation treatments and rliable-based sta…

    Python

  3. overleaf-mcp overleaf-mcp Public

    Edit your Overleaf projects from Claude. Local, auditable, single-user by design. stdio for Claude Desktop, HTTP for claude.ai web.

    Python

  4. claude-teams-operator claude-teams-operator Public

    Kubernetes operator that runs Claude Code Agent Teams as distributed pods. Supports coding and Cowork modes for knowledge work.

    Go

  5. paper-skills paper-skills Public

    Claude Code skills for academic paper triage, annotation, and Obsidian vault integration. Screen papers by relevance, generate YAML frontmatter annotations, and run a unified intake pipeline.

    1

  6. engineering-handbook engineering-handbook Public

    How I build software, and the philosophies behind it

    Shell