Speaker

Michael Limberger

Senior Web & Linux Administrator

Faculty of Dentistry, University of Toronto

Michael builds and maintains web applications, Linux servers, and AI systems for the University of Toronto's Faculty of Dentistry. With over 15 years of Perl expertise, he has a track record of solving real-world problems with practical, maintainable code.

He built the RAG Proxy system presented at TKF 2026: a local AI layer that searches 94 dental policy documents and generates cited answers without sending a single byte of data to the cloud. The entire system runs on one Mac Studio using open-source tools.

Outside of work, Michael runs Techalicious-Club, a GTA meetup group for tech hobbyists covering AI tools, Linux, privacy tech, and self-hosting. He is a strong advocate for sovereign AI: local, private, and free from vendor dependency. In his view, healthcare IT needs to own its intelligence, not rent it.

Personal Projects

Beyond the day job, Michael builds tools and systems that scratch his own itch. These are personal projects, built on evenings and weekends, driven by curiosity about how AI, systems programming, and language design can work better together.

TOAD

Total Orchestration And Dispatch, a symbolic systems language

A programming language Michael designed and built from scratch in Rust. TOAD uses zero English keywords for logic. Every control structure is a symbol. It's a conductor language: it handles variables, control flow, regex, and system execution while delegating heavy lifting (networking, parsing, databases) to external programs via a polyglot bridge system. TOAD scripts can embed Perl, Ruby, and Python inline, passing typed data between languages through a JSON bridge. Compiles to a single static binary with no runtime dependencies.

Rust v0.11.0 1,348+ tests 28 test files ~9,500 lines

System Commander

AI-powered sysadmin agent, heuristics first, LLM second

An AI-powered sysadmin agent built in Perl/Mojolicious. Uses a 5-layer heuristic decision tree (regex pipeline) as its primary brain, with a local LLM (Ollama) as a fallback normalizer for ambiguous inputs. Includes a semantic memory system (Brain.pm) with SQLite + PDL vector search, the same engine that powers the RAG Proxy. The latest version (v3) integrates the brain CLI for persistent knowledge across sessions.

Perl Mojolicious 12 HDT modules 1,916 tests Brain.pm

Battle-Grounds

Top-down 2D team PvP brawler with fantasy combat and six character classes

A standalone 4v4 team deathmatch game built in Godot 4.6 with GDScript. Two teams (Blue vs Yellow) battle across a map populated by hostile NPCs. Six character classes (Demon, Ent, Lich, Mushroom, Beholder, and Lizardman), each with team-specific special abilities (burn vs ice slow, lifesteal vs zombie spawn, dodge vs double attack). Neutral Soldiers and Champions roam the map dropping healing potions, power potions, and thunder wands. First team to 25 kills wins. CPU backfills empty slots with bot AI. LAN multiplayer planned.

Godot 4.6 GDScript 6 classes 4v4 PvP MVP complete

View the Presentation →