hierarchical-agents

A multi-agent research laboratory for data distillation and long evolutionary collaborative reasoning.

Master thesis work: explaining deep reinforcement learning policies

Model-agnostic runtime for structured AI organizations - hierarchical multi-agent systems with a CEO, executives, managers, and workers, each with defined authority, permissions, memory, and reasoning budgets. Quota-aware routing across local and frontier models. You talk to the CEO; the organization does the rest.

A Claude Code skill that builds hierarchical, domain-agnostic agent systems from a single objective, via a central Designator agent. Self-building, cost-aware, model-agnostic.

Hierarchical agent orchestration for OpenCode: specialized roles at every scope, budgeted context, parallel builders, evidence-gated completion and a native desktop console.

Autonomous engineering control plane: a five-tier hierarchical multi-agent DAG (Supervisor > Managers > Workers > Validator > Reducer) that follows software changes from PR to production - evidence-driven, provenance-preserving, spec-driven.

Parallel Mind 2.0 ; A hierarchical agent system for parallel task execution.

Advanced CrewAI: learning path generator (SerperDev + custom tool) + code debugger (CodeInterpreterTool)

Orchestrate hierarchical agents in OpenCode with dependency-aware boards, parallel builders, and evidence-gated completion.

A full-stack AI research agent — three-agent pipeline (plan, search, synthesize) with live source tracking, search depth control, and document export. Originally prototyped in n8n + Replit, rebuilt as a standalone Next.js + FastAPI app with an MCP server tool layer.

AI-assisted Pokemon player research: learned goal selection, verified Red collection, deterministic skills and a shared Pokedex. Active development.