Skip to main content
Bogdan CozmaBC
Open to opportunities

Bogdan Cozma

@bogdancozma

I design adversarial AI evaluations, rubrics, and agent scenarios that expose model failures.

Romania
Message

At StellarAI, I designed end-to-end evaluation scenarios for LLMs, including prompts, tool-call selection, realistic databases, and gold-standard trajectories targeted at an 80%+ model failure rate.

I create coding prompts across C++, Python, JavaScript, and Bash, then write multi-tier rubrics to compare outputs from competing models on correctness, completeness, readability, and task alignment. I also build virtual-assistant simulations, annotate cross-OS navigation tasks, review peer submissions, and make final accept/reject decisions.

I bring a Python and software QA background, including a seven-stage astrometry pipeline that achieved 92.7% completeness with 0% false positives on a synthetic 274-source image.

Experience

Work history, roles, and key accomplishments

ST

AI Trainer & Evaluator

StellarAI

Nov 2024 - Jun 2026 (1 year 7 months)

Designed end-to-end evaluation scenarios for LLM training, including system/user prompts, tool-call selection, and realistic databases, targeting at least an 80% model failure rate. Performed QA on AI-generated code and authored multi-tier rubrics to evaluate outputs from competing models.

Education

Degrees, certifications, and relevant coursework

HU

Hyperion University

Bachelor's Degree, Automation and Applied Informatics

2022 - 2026

Grade: 9.50/10

Pursuing a degree in Automation and Applied Informatics, with a final project on Python module for automatic identification of objects in photometric images.

Tech stack

Software and tools used professionally

Get matched with your dream remote job

Sign up now and join over 250,000+ remote workers who receive personalized job alerts, curated job matches, and more for free!

Sign up
Himalayas profile for an example user named Frankie Sullivan