Lab

AI-powered League of Legends companion — runes, items and augment recommendations in real time

Connects to the League client over the LCU WebSocket, profiles all ten champions in a draft (role, playstyle, damage profile, CC), and auto-imports optimized rune pages during champion select. A three-tier pipeline: rule-based recommendations, data-driven predictions on Diamond+ matches, and Claude-generated explanations of why a choice wins this specific matchup.

highlightAn ARAM interaction engine with 71+ tagged mechanic chains that detects 'broken' combos mechanically (infinite-scaling scenarios statistics miss) — verified by a golden no-regression test suite of 120+ tests.

stack: React · FastAPI · Python · Electron · Claude API · SQLite

Real-time in-game coaching overlay with voice — a Challenger-level coach powered by Claude

A transparent Electron overlay rendered on top of a running DirectX game, toggled with a global hotkey. A Python service monitors live game state (gold diff, objectives, KDA) and streams coaching tips over WebSocket; post-game it produces a graded review with timestamped mistakes. Voice coaching via TTS, with Spanish localization.

highlightSolving overlay-on-game rendering plus real-time Claude prompting with structured Pydantic schemas for matchup analysis, jungle prediction heatmaps, and S–D graded post-game reports.

stack: Electron · React · FastAPI · Claude API · WebSocket · TTS

Multi-device persistent memory for Claude as an MCP server

An MCP server that gives Claude durable, searchable memory shared across devices: facts stored in Turso (distributed SQLite) with FTS5 full-text search, exposed as tools any MCP-capable client can call.

highlightMemory recall as ranked full-text search over a synced edge database — the same pattern this portfolio uses to stay readable by agents.

stack: MCP · TypeScript · Turso · SQLite FTS5

MCP servers that do release paperwork — ServiceNow extraction and Azure DevOps change requests

Custom MCP servers used in real release cycles: one drives ServiceNow through Playwright to extract ticket data; another generates Change Request artifacts and CAB material from Azure DevOps work items. Plus OpenClaw, a game-automation agent driven by natural-language commands, built to stress-test the same agent patterns.

highlightNot demos: these run during actual quarterly releases and save hours of change-management paperwork per cycle.

stack: MCP · Python · Playwright · Azure DevOps