AI Engineer · Applied AI / LLM · AEC
I build AI systems for document-heavy, domain-specific work.
I started as an architect. That background still shapes how I engineer software: understand the workflow first, make outputs traceable, and ship something people can actually use.
Today I build ArchNorm, a B2B product for construction-document compliance, and work on AI for landscape and site design. I am also pursuing a master's degree in Artificial Intelligence at ITMO University's AI Talent Hub.
ArchNorm answers questions against construction regulations and points back to the source. I built it from problem discovery and system architecture through the Python/FastAPI backend, hybrid retrieval, document processing, evaluation, and local/on-prem deployment.
The production code and regulatory corpus are private. The public preview shows the product without exposing customer data or the proprietary pipeline.
A deterministic routing engine that distributes payouts across providers under hard eligibility rules, capacity limits, conversion estimates, cost constraints, and fallback scenarios. Every decision is traceable.
Implemented in Ruby 3.3 with the standard library only: 71 tests, 503 assertions, reproducible seeded runs, an interactive review stand, and documented trade-offs.
An open lab for AI-native tools in architecture, 3D, games and creative work: a catalog of tools, discussions and critique, and contributions to the open projects people already use.
vibecading.io · Catalog · Discussions
- RustGIS: geometric interval and standard deviation classes for graduated colours (#805, merged); manual class breaks (#822)
- BIMCraft: Windows start fallback to DirectX 12 (#63); a Type Selector list that fits the window (#67)
- build123d: a regression test for STEP profiles placed in a sketch (#1546)
- RhinoAI (McNeel): decimal tool parameters through the MCP router (#207)
A security testing toolkit for AI agents, built with a team for an agentic red-teaming hackathon. Repository
Python, Rust, Go, FastAPI, RAG evaluation, Document AI, tool-using agents, guardrails, and local inference. For AEC workflows: Rhinoceros, Grasshopper, Revit, Archicad, and geospatial data.