About
I build AI systems for construction companies. My flagship project Green manages a live HVAC crew's daily operations — 12+ AI agents, 30+ automated tasks, 12,000+ pages of building codes searchable by voice. Not a demo. Production.
Why it works where other construction AI fails: I work the jobsite every day. I'm a union sheet metal worker with Sheet Metal Workers Local 66 in Everett, WA. My users are my coworkers. My testing ground is the jobsite I walk onto every morning. I don't guess what field crews need — I am the field crew.
20+ years in construction and engineering — from a civil engineering degree and running manufacturing businesses in Russia, to immigrating to the US and working through every layer of HVAC: gas pipe, equipment install, TAB, calibration, commercial sheet metal. That's the foundation behind every AI system I build.
I also build multi-agent AI architectures and custom trading tools — 21-entity trading OS, 15,000+ lines of C# for NinjaTrader. If it needs to survive real conditions and real users, I know how to build it.
Services
AI Field Intelligence
Voice-first AI systems for construction crews. Code lookup, field support, automated operations — all through Telegram. Like Green, built for your company.
Custom AI Automation
Multi-agent AI systems built around your workflows. From architecture to deployment. Production-grade, not prototypes.
AI Audit & Strategy
Already have AI tools that underperform? I audit what you have, identify gaps, and build a roadmap that works in real conditions.
Work
-
Green Flagship
Problem: Construction crews waste 35% of time searching for answers across 12,000+ pages of building codes.
Solution: AI field intelligence platform with 12+ specialized agents, voice-first access, and automated daily operations.
Result: Deployed with a live HVAC crew. 30+ daily tasks automated. Every answer cites the exact code section.
- AI Platform
- Construction
- Voice-First
- Claude API
- Python
-
Meridian
Problem: Futures traders lose edge to fatigue, fragmented tools, and inconsistent review.
Solution: 21-entity modular AI operating system — journal, risk management, psychology tracking, discipline enforcement.
Result: 258 files, ~33K lines of documentation. Single Telegram entry point. File-based inspectable memory.
- AI Architecture
- Multi-Agent
- Claude Code
- Telegram
-
NinjaTrader Trading Suite
Problem: Retail traders lack institutional-caliber tools for structured execution and risk management.
Solution: 8 custom products — strategies, indicators, analysis tools. Unified signal protocol and R-based risk framework.
Result: 15,000+ lines of C#. 200+ configurable parameters. Production-deployed.
- C#
- NinjaScript
- NinjaTrader 8
- SharpDX