Building tools for AI agents
Deterministic engines that let AI agents act on real state—and then check the work. The model proposes. The engine validates.
Tools don't think, they act.
Every engine follows the same contract: Deterministic, Idempotent, Auditable, Composable, Verifier-gated. The LLM proposes. The engine validates. State comes from the database, not hallucination.
Building at Scale
Core Projects
Infrastructure that lets AI agents act on real state — and verifies the result before anyone trusts it.
Multi-Agent Framework
Production-ready multi-agent coordination with TDD workflows (Red→Green→Blue). Role contracts for Orchestrator, Architect, Planner, Code, Debug. IDE integrations for Roo Code, Claude Code, Cursor.
RPG MCP
Full D&D 5e backend for AI Dungeon Masters — combat, spells, quests, conditions and spatial calculations, with 1100+ creature presets. Powers Quest Keeper AI.
Quest Keeper AI
Desktop RPG companion (Tauri) combining an AI Dungeon Master with a visual game engine. 3D battlemap, 2D world map, spellcasting, NPC memory — on a 7-layer context budget.
LLM-Driven Proof Search
An evidence-preserving environment for AI-assisted mathematical research: formalization, decomposition, certificate generation, Lean verification and publication export.
OODA MCP
Full computer control implementing the OODA pattern — observe (screen), orient (context), decide (LLM), act (input). 62 tools for desktop automation.
NLS — Natural Language Source
The missing middle layer between AI conversation and production code: structured English that compiles deterministically to Python.
More Projects
Formal verification, research tooling and binary analysis.
remcp — Reverse Engineering MCP
Read-only binary analysis for agents: PE/ELF/Mach-O/JVM/.NET, disassembly, cross-build matching, a verified symbol knowledge base, JVM decompilation and crypto identification.
Chess Vision Studio
See the forces on the board, not just the evaluation — a 2D chess perception engine with validated overlays, plain-language coaching and insights across your game history.
Prompt Engineering Taxonomy
Interactive taxonomy of 182 prompt-engineering techniques from 20+ papers — searchable, categorised, with visualisations and JSON export.
Want to learn more?
Check out my GitHub for full documentation, source code, and live demos of all projects.