HG Stack
An independently developed project building AI-agent infrastructure, reproducible evaluation workflows, automated research pipelines and systematic market tools.
Agent systems
Provider-agnostic workflows with explicit cost controls, deterministic fallbacks and fail-safe execution.
Benchmarks
Reproducible evaluation pipelines for agent benchmarks, including CHI-Bench experiments and official judge workflows.
Market research
Read-only market-data pipelines, ranking models, opportunity discovery and forward-testing infrastructure.
Automation
Event-driven and scheduled routines across cloud infrastructure, GitHub Actions and durable state.
How Claude is used
Claude is used in evaluation workflows where Anthropic models are required, including benchmark judge execution, and is being evaluated for broader agent reasoning and benchmark workloads. The architecture remains multi-provider by design.
Stage
HG Stack is bootstrapped, has no external funding, and is currently in active technical development and experimentation.