Daniel Culotta

Designed and built a physics-informed predictive maintenance system for industrial bearing fault detection using a proprietary Multi-Scale CNN + Temporal Self-Attention + Recursive Reasoner architecture Created the Physics Validator: an explainability module that verifies AI predictions using harmonic frequencies (BPFO/BPFI) and modulation sidebands — matching expert vibration analyst methodology Achieved % false alarm rate (industry average: 25-35%), 14.8ms inference on CPU, 2.7MB model size

Validated on NASA C-MAPSS, CWRU, and Paderborn benchmarks with 94.5% transfer learning accuracy Developed B2B SaaS model

Built a B2B SaaS platform for EU regulatory compliance (CSRD, AI Act, DORA, NIS2, CSDDD): 12,300 lines Python + 2,850 lines TypeScript Implemented hybrid RAG pipeline (BGE-base dense + BM25 sparse + Reciprocal Rank Fusion) with legal-aware contextual chunking inspired by Anthropic research Built LangGraph multi-agent system: Q&A Bot, Gap Analyst, Regulatory Monitor with async orchestration Direct EUR-Lex SPARQL/CELLAR integration for real-time regulatory monitoring from official EU sources Evaluation framework (Monte Carlo + LLM-as-Judge + Golden Set) scored 9.0/10 by independent 6-agent audit Stack: FastAPI, SQLAlchemy 2.0, Pydantic v2, PostgreSQL, Qdrant, Redis, Next.js 14, Docker with security hardening 159 unit tests, GitHub Actions CI/CD, multi-tenant architecture with Row-Level Security