AI Engineer building production LLM agents, retrieval systems, evaluation pipelines, and multilingual AI.
I build AI systems that need to work outside controlled demos: tool-using agents, voice workflows, source-grounded research systems, model evaluations, and low-latency LLM services.
Professionally, I have worked across real estate, financial research, healthcare, and education:
- Built a Russian real-estate voice agent with sub-250 ms p95 latency while reducing infrastructure cost by ~35%.
- Automated financial research workflows, reducing analyst effort by ~60% and saving 4+ hours per report.
- Developed EEG analysis and clinical-support tools that reduced query-resolution time by ~50%.
Right now, I’m working on:
- natural, conversational Persian TTS;
- multilingual OCR and VLM evaluation;
- evidence-driven research agents with citations and validation gates;
- post-training and evaluating small open models.
I also write LLM Engineering in Action and the Building Production AI newsletter.
Evals over vibes. Evidence over confident output. Measure latency, cost, quality, and failure modes.