Led team scaling from single dev to 3-person fullstack team. Established CI/CD practices, mentored members, defined roadmap. Made critical Azure API pricing/performance decisions. Handles 100+ RAG configurations, 200k daily requests.
Jan Czechowski legal: Przemyslaw Jan Czechowski
Selected Projects
Built interactive ELO scoring dashboard to evaluate and compare GPT, Claude, Gemini and other models based on human agents' selection preferences from multiple model suggestions. Developed end-to-end fine-tuning (GPT-3.5/4o) and evaluation pipeline. Designed strategy for systematic model rollouts based on real-time (daily) preference data from agents. Built API serving 4000+ suggestions daily from diverse models/providers.
Designed and built ensemble spellchecker with custom seq2seq transformer, owning the complete roadmap: dataset engineering, model training, evaluation, deployment, monitoring, and KPI definition. Delivered production system: 10+ req/s at 200ms median response time, <1% overcorrection rate. Revisited twice over 18 months for retraining and optimization.