04About
About
I'm an AI and software engineer in Boston, MA. At Hime I build Wheelbase end to end, from Postgres security policies to the retrieval layer to the React app dealers use every day. On my own I built Grammario and Lociros, NLP products for language learners, and at the Massachusetts Executive Office for Administration and Finance I built AWS Bedrock pipelines that translate financial documents without breaking their layout.
Most of my work spans the stack: database schemas and their security policies, search and retrieval, model calls, APIs, and the interface on top. Two changes I was happy with: rewriting Wheelbase's backend in Go cut database load by more than 80%, and running Grammario's analysis steps in parallel brought a nine-second wait down to about four.
With AI features I try to get the structure right before involving a model. In Grammario a parser finds the grammar and the LLM only explains it. In Lociros a morphological analyzer checks each generated passage against its level. In Wheelbase the assistant can write SQL, but only inside a restricted Postgres function, and destructive changes need approval. Models are useful but make mistakes, so I add checks where a mistake would matter.
When Wheelbase ingestion got too slow in Python, I learned Go and rewrote it. I also fixed a small pandas plotting bug that was merged upstream. Outside of code I read history, science, and philosophy, and I'm usually partway through learning another language.
I — Languages
Languages
I speak Russian and I'm learning Italian, Turkish, German, Hebrew, and Japanese, which is a big part of why I got into NLP. Turkish builds words by stacking suffixes while Italian inflects them, so Grammario uses a different analysis approach for each language family.
English
English · Native
Русский
Russian · Native
Italiano
Italian · Proficient
Türkçe
Turkish · Proficient
Deutsch
German · Basic
עברית
Hebrew · Basic
日本語
Japanese · Basic
II — Toolkit
Skills
- Python
- Go
- SQL
- TypeScript
- C / C++
- Bash
- R
AI & LLM systems
Production RAG ranking at Wheelbase, eight structured-output LLM services in Grammario, governed natural-language SQL, AWS Bedrock pipelines in government.
RAG · Hybrid retrieval · Embedding pipelines · Prompt engineering · Structured JSON outputs · OpenRouter · OpenAI · Anthropic Claude · AWS Bedrock
NLP
Dependency parsing and morphology across six languages, CEFR scoring from engineered features, analyzer-validated LLM generation.
spaCy · Stanza · sentence-transformers · Sudachi · pymorphy3 · CAMeL Tools · NLTK · Universal Dependencies
Machine learning & data
MLflow-tracked models served behind FastAPI, CNNs exported to Core ML, pipelines over 177K+ records and streaming ETL.
PyTorch · TensorFlow · scikit-learn · MLflow · pandas · NumPy · Core ML · ETL / streaming ingestion
Backend & databases
A Go service that cut DB load 80%+, a 66-table RLS schema, pgvector search, Redis caching layers.
Go / Gin · Python / FastAPI · PostgreSQL · Supabase · pgvector · Redis · SQLite · tRPC · REST
Frontend & apps
A 68-route React 19 app, an Electron desktop shell running tRPC over IPC, an Expo field app with a custom native VIN scanner.
TypeScript · React · Next.js · TanStack · Tailwind CSS · Electron · Expo / React Native · Yjs · ReactFlow
Infrastructure
Per-service Docker images behind Nginx, path-filtered GitHub Actions deploys, serverless background jobs on Vercel.
Docker · Nginx · GitHub Actions · AWS · Vercel · DigitalOcean · Dokploy · MinIO · Stripe
III — Contact
Write to me
Open to roles in AI engineering, ML, data, and software engineering, Boston metro or US-remote. US citizen, no sponsorship needed.