AIDA
AI Document Analysis: uploaded contracts, policies and statements run through OCR and a staged LLM pipeline that writes structured, audited data onto real CRM and asset records.
- Python
- Django
- Pydantic
- LiteLLM
- Azure AI Foundry
- GDPR
Hi, I'm Cezar. AI Engineer.
AI Document Analysis: uploaded contracts, policies and statements run through OCR and a staged LLM pipeline that writes structured, audited data onto real CRM and asset records.
Video coaching tool that breaks each attempt into phases and grades technique against exact frame references.
On-edge multi-camera detection driving an 8-point directional audio deterrent across a fish farm, 24/7.
AI-powered trash identification and sorting bin: computer vision on edge hardware to classify waste in real time.
Bachelor thesis. Researched whether stem-aware neural networks can match human-made masters in metric and perceived accuracy. Short answer: artistry still wins, the human ear catches what the model misses.

I was building things that had to fly before I could write code: model aircraft and gunpowder rockets, built from raw materials and judged in competition.
That instinct never left. Eight years of formal percussion training, 50+ aeromodelling competitions and 35+ medals, then four years of STEP IT Academy from age 12: programming, hardware, IoT. The habit underneath all of it is the same. Build the whole thing, then test it against reality.
At Howest (Creative Technologies & AI, Belgium) every semester ended with shipping to a real company. The last of those, an agentic system for a firm processing financial documents, became my job. AI Square Group hired me off the delivery, where I built the engine that detects and pseudonymises personal data, live in production under GDPR. I work spec-first, with verification loops and exit criteria. Agentic engineering, not vibe coding.
The range runs wider than pipelines. A Mitacs-funded research internship at Algoma University on quantum-enhanced recommender systems using D-Wave annealers, and a thesis testing whether stem-aware AI mastering can match a human engineer. It can’t yet, and the reason says more about how we build software than about audio. What I’m after is owning outcomes end to end and talking directly to the people the system is built for.
35+ medals, aviation & rocketry · 8 years percussion · 4 years STEP IT Academy