Maxim Thibodeau | AI Systems Architect

AI Systems Architect • Macro-Agent Builder

Web Access • Tool Use • Verification • Retro-Action • Systems Engineering

I build connected AI systems that combine web interfaces, backend services, database integration, robotics control, simulation, and agent-style automation.

My current focus is a basic AI macro-agent architecture: an AI model receives a goal, searches or reads context, plans actions, executes safe steps, verifies the result, and improves through retro-action.

Open Agent Page View CV Hardware / Low-Level

Featured Project: AI Macro Agent

This project describes a practical agent loop for companies building AI agents around models, tools, browser/web access, APIs, and feedback.

1. Observe Read the user goal, files, logs, pages, API responses, and system state.

2. Plan Break the goal into safe, testable, reversible actions.

3. Act Use tools such as web search, API calls, patches, SQL, or command execution.

4. Verify Check status codes, tests, syntax, data, and user-visible evidence.

5. Retro-Action Feed errors and observations back into the next cycle.

See Full Agent Architecture

Agent Capabilities

Web-aware research before decision-making.
Tool calling through controlled APIs and backend routes.
Patch generation for PHP, Python, JavaScript, SQL, and server config.
Verification using logs, curl tests, syntax checks, and rollback backups.
Retro-action loop that corrects failed actions instead of blindly continuing.

System Stack

Frontend: PHP, HTML, CSS, JavaScript.
Backend: FastAPI, Uvicorn, Nginx, API routing.
Data: MariaDB integration for real platform data.
AI: model prompt layer, tool routing, verification, feedback.
Robotics: AI command interface, DOF planning, simulation pipeline.

Integrated Project Network

My website connects multiple real systems into one portfolio: an AI Agent page, my CV, and the Hardware / Low-Level page.

Agent CV Hardware

Hardware / Low-Level

Restored hardware demo page with Mini OS and Advanced OS video sections, plus a low-level C scheduler sketch built around timer interrupts and task flags.

Open Hardware

Why This Matters

AI agents need more than text generation. They need access to context, controlled action, memory, validation, recovery, and a clear interface for human review. My work shows this direction through real web systems, deployment patches, database integration, and robotics/agent experiments.