AI & Data Engineer · Montréal
I'm Oleksandr. I build AI and data systems that hold up beyond the demo — with care for reliability, evaluation, and the people using them.
From physics to production.
01 / Selected projects
Things I've been building.
relai.
Your AI coding sessions in one workspace. Resume in the built-in terminal, carry context between tools, and track spend.
Developer tools macOSgramp.
Your genome, health records, and research in one encrypted workspace — with illustrated anatomy and simple summaries.
Personal health In development
Brosophagus.
Check food labels and ingredients against your personal triggers, then track how you actually felt.
Food & health iOS
CôtéRue.
Decode Montréal parking signs, save your spot, and get reminders before it's time to move.
City life iOS & Android02 / What I work on
From the model
to the messy reality.
I build large-scale data and machine learning systems, with a recent focus on GenAI, RAG pipelines, and agentic AI. My work spans healthcare, ad-tech, and e-commerce.
I care most about the unglamorous parts of real-world AI — evaluation, reliability, cost, and what happens when the model confidently gets it wrong.
- Agentic AI: multi-agent orchestration (LangChain/LangGraph, CrewAI, Pydantic AI, Bedrock AgentCore), tool-use design, failure-mode handling, human-in-the-loop.
- LLMs & RAG: Anthropic, OpenAI, Llama, Mistral; hybrid retrieval (OpenSearch, pgvector); guardrails, grounding, ranking; local inference with Ollama, llama.cpp, vLLM.
- Evals & reliability: LangSmith, LLM-as-judge, golden datasets, CI/CD eval gating.
- MCP: internal Model Context Protocol servers with auth, scoped tools, streaming, and schema validation.
- Platform: Python (FastAPI, async), AWS (Bedrock, HealthOmics, Lambda, Step Functions, Redshift, S3, DynamoDB), Airflow, Docker, Kubernetes.
03 / Background
A physicist's curiosity.
An engineer's approach.
I've spent the last several years building production GenAI systems — large-scale RAG pipelines for clinical data, MCP servers for text-to-SQL flows, multi-agent workflows, and the eval infrastructure around them. Earlier work was in ML and data engineering at Gazelle AI, SSENSE, Bandsintown, and Wajam, and I taught a Big Data course at Concordia University.
PhD in Physics (All But Defence) from the University at Albany, SUNY — dissertation combined stochastic Liouville equation modeling for NMR with supervised machine learning on Raman spectroscopy data. The full history is on the Work page.
Follow my career04 / Research
Published work.
- Earl K., Kazakov O. (2020). The Spin Echo, Entropy, and Experimental Design.Proceedings, 33(1), 34.doi
- Ryzhikova E., Sikirzhytski V., Kazakov O., Halamkova L., Quinn J., Zimmerman E.A., Lednev I.K. (2016). Raman spectroscopy of Cerebrospinal Fluid for Alzheimer's disease diagnosis.Journal of Biophotonics.
- Ryzhikova E., Kazakov O., Halamkova L., Celmins D., Malone P., Molho E., Zimmerman E.A., Lednev I.K. (2015). Raman spectroscopy of blood serum for Alzheimer's disease diagnostics: specificity relative to other types of dementia.Journal of Biophotonics, 8(7), 584–596.doi
- Lednev I.K., Ryzhikova E., Kazakov O., Halamkova L., Celmins D., Malone P., Molho E., Zimmerman E.A. (2014). Raman Spectroscopy of Blood for Alzheimer's Disease Diagnostics.Annals of Neurology, 76, 94.
05 / Say hello
Something interesting
on your mind?
AI engineering, a project collaboration, or just a good conversation. Send a short note about what you're building and where you need help.