Sebastian Buzdugan

📍 Cluj-Napoca  |  🧪 ML Engineer + PhD Student in AI  |  🛡️ Building open-source Responsible-AI tooling

I build tools that make AI safer to ship and easier to learn. Creator of the FRAI responsible-AI toolkit, and I teach modern LLM and agent engineering.

---

🛡️ Responsible AI (FRAI)

• frai: Open-source toolkit for responsible AI. CLI + SDK to scan code, collect evidence, and generate model cards, risk files, evals, and RAG indexes.
• frai-skills: Agent skills for responsible AI. The FRAI Gate makes AI coding agents answer 7 responsible-AI checks in every spec, with npx frai-gate blocking CI until they are. Install: npx skills add sebuzdugan/frai-skills.
• frai-benchmark: A consensus "Panel of Experts" safety and compliance benchmark for SOTA LLMs (DeepSeek, Grok, GPT-5).
• frai-chat: Responsible-AI copilot grounded in the EU AI Act, NIST AI RMF, and ISO/IEC 42001. BYOK, runs in the browser.
• frai-judge: Model-as-judge evaluation used in the FRAI stack.
• framework-rai-mcp: An MCP server that brings responsible-AI checks to agents.

More at frai.cc.

🛠️ Practical tools

• CrossCheckAI: Ask multiple frontier LLMs the same question and see where they agree, disagree, and why the disagreement matters. Web + CLI + library, BYOK, runs in the browser. Live demo.
• AIEngineerCV: A CV builder for AI engineers. Web + CLI + Claude plugin, with an AI Recruiter Score.
• frontier-ai-prompts: A practical prompt library for GPT-5.5, Claude Opus, Gemini, and modern AI workflows.
• async-nlp-pipeline: A production-style FastAPI NLP service: async preprocessing, transformer sentiment, spaCy NER, rate limiting, SQLModel.
• chatbot-assessment: An AI chatbot for appointment scheduling.

📚 Learning & research

• langchain-course-2026: 10 Days of LangChain. Build an agentic learning assistant with LangGraph and RAG.
• LLM: Hands-on LLM fine-tuning and RAG: Gemma and LLaMA-2 (LoRA and full FT), Hugging Face workflows, end-to-end prototypes.
• gpt-v2: GPT-style NLP from scratch: data extraction, bigram language modeling, custom training, early chatbot prototypes.
• multi-agent-system: A multi-agent simulation where heterogeneous agents coordinate under partial observability.
• MRIimagepreprocessing: MRI image preprocessing for medical-imaging ML.
• audio-fingerprint: Audio fingerprinting in Python.

_{Plus neural-network coursework and earlier full-stack / Flutter projects across my repositories.}

🤝 Connect

_{Open source, responsible by default. Building in public from Cluj-Napoca.}