LLM integrations
OpenAI, Anthropic, Gemini, open-weight models via OpenRouter. Structured outputs, function calling, context management, cost controls and provider fallback routing.
AI/ML Engineer · research systems
Pavel
Kazantsev
I build experiments that can show their work.
I build AI products, evaluation systems and data-intensive experiments, backed by seven years of production software engineering. I publish the architecture, evidence and failures.
Writing
Engineering what survives
My Football Model Passed Validation. A Five-Check Audit Killed It.
Val away CLV was +152 bps, confidence interval excluded zero. Then I ran a five-check forensic audit: opening-line baseline, Brier skill, feature contamination, incremental CLV, and season stability. Four failed.
Read the case study →What I work on
AI beyond the demo
The model is one component. The product also needs data, orchestration, evaluation, interfaces and failure handling.
NLP & RAG systems
Retrieval-augmented generation, embedding pipelines, reranking, chunking strategies and structured output extraction — built for production accuracy, not demo conditions.
ML pipelines & research
Walk-forward OOS, leakage controls, point-in-time data, cost models and immutable experiment artifacts. Results that can be interrogated, not just reported.
Agentic workflows
Typed tools, resumable stages, deterministic evaluation gates and explicit human approval — instead of autonomous systems with hidden state.