walkerhughes/terminal-bench
Measuring and evolving with the frontier of agent work
math, ml, bjj
Repositories
walkerhughes/terminal-bench-ci-e2e-1793
Temporary end-to-end validation of terminal-bench cross-platform CI
walkerhughes/agents
Helpful tools for working with Claude Code & Codex.
walkerhughes/tastytrade-mcp
MCP built on the TastyTrade OpenAPI spec. Let agents access to your brokerage account, market data (options chains, prices, indicators), and order management for options trading.
walkerhughes/mcp-auth
OAuth to a MCP server distributed as a Claude Code plugin.
walkerhughes/harbor
Harbor is a framework for running agent evaluations and creating and using RL environments.
walkerhughes/rag
walkerhughes/mctsr
Monte Carlo Tree Self-refine implementation for enhanced LLM accuracy in Q&A tasks based on MCTSr paper, based on Trelis Research implementation.
walkerhughes/harbor-mcp
MCP server to interact with the Harbor Hub.
walkerhughes/dotfiles
walkerhughes/t-bench
A benchmark for LLMs on complicated tasks in the terminal
walkerhughes/pydantic-ai-skills
This package implements Agent Skills (https://agentskills.io) support with progressive disclosure for Pydantic AI. Supports filesystem and programmatic skills.
walkerhughes/evals-in-ci
walkerhughes/llm-sft-grpo
Teaching LLM agents to play (easy) games via multi-turn SFT + GRPO training. Increased win-rate by 47% over baseline on 0.5B param model.
walkerhughes/agent-sandboxing
Sandboxed agent with human-in-the-loop checkpointing. Containers exit when waiting for user input and resume with full context for $0 in idle container costs.
walkerhughes/rpg_svo_pro_open
walkerhughes/multi-turn-evals
Evals for both single-step and full-trajectory traces in multi-turn agent workflow.
walkerhughes/claude-code-uv-template
Template repo for using uv and Claude Code for development. Includes pytest, ruff, mypy, coverage, CI, and pre-commit hooks.
walkerhughes/agent-tool-use-evals
walkerhughes/deep-research-sandbox
walkerhughes/multistage-retrieval-pipeline
Full-stack hybrid RAG application with keyword search (Postgres FTS), semantic re-ranking (pgvector), and evals in CI. Chat about episodes of the Dwarkesh Patel podcast.
walkerhughes/walkerhughes
Config files for my GitHub profile.
walkerhughes/mlops
walkerhughes/yield-curve
Daily AI-generated newsletter on macroeconomic environment and major trends. Automates data ingestion pipeline via GitHub Actions and stores data in Google BigQuery [not maintained].
walkerhughes/debate_duel
An arena for head-to-head debates between teams of AI agents on hot topics. Agents plan, research, and present their arguments to an LLM Judge, determining a winner and updating teams' ELO rankings.
walkerhughes/slideagent
walkerhughes/sentiment
repo for MSDS697 Distributed Data Systems
walkerhughes/longitudinal
walkerhughes/Autoencoders
Exploration of data generation techniques with Variational Autoencoders in PyTorch, incorporating custom loss functions and deploying decoder with GCP.
walkerhughes/yc-pypi
Python package published to PyPI for quantitative yield curve analysis.