Summary
- 20+ years in software development: C++, system programming, 3D/2D graphics, AI infrastructure
- AI Agents: architecture of autonomous orchestrators for CAD/EDA with native C++ implementation (HTTP, JSON, function-calling)
- Low-level AI: custom LLM communication protocols, context optimization, multi-GPU inference (RTX 4090+5090)
- 3D/2D Graphics: OpenGL, shaders, SDF, real-time rendering of million-element datasets
- Enterprise-ready: on-premise AI, privacy-first architecture, cross-platform (Windows/Linux)
- Leadership: LLM integration initiatives, code review, mentoring, stakeholder communication
- Planning: resource estimation, timeline management, task prioritization under budget constraints
- Process establishment: code review, CI/CD, distributed debugging, performance metrics collection
- Graphics expertise: OpenGL, DirectX, shaders, Signed Distance Functions
- Privacy-First AI: local LLM deployment (Ollama, Qwen) on RTX 4090+5090 (56GB VRAM) for sensitive data
Work Experience
Electromagnetic Signal Integrity CAD Software for high-speed PCB design
Senior Software Developer | AI Agent & Graphics Lead | 2015 – Present
AI orchestrator Simbeor Commander (LLM, function-calling), OpenGL 3D/2D GUI, SDK (Python/Matlab/Altium), Linux port, distributed agents.
Game Development / Multiplayer
Lead Software Developer | 2014
Game server, deferred renderer, flight physics, remote diagnostics, 3ds Max pipeline integration.
3D Printing / Computational Geometry
Software Developer | 2012
Mesh healing, hole filling, decimation algorithms for STL processing.
Ancient Gate / Il-2 Sturmovik: Birds of Prey (Nintendo DS)
(Amazon)Console Game Development
Lead Software Developer | 2009 – 2010
DS emulation layer, fixed-point ARM physics, dynamic LOD rendering.
Tech Stack
AI & LLM Infrastructure
Agentic Workflows, Function Calling, Custom HTTP/TCP (C++), JSON Parsing, Context Optimization, Local Inference (Ollama/vLLM), Multi-Provider Integration, Privacy-First Architecture
System Programming
C++ (11/14/17/20), STL, Boost, Linux/Windows, POSIX, WinAPI, Multi-threading, Distributed Computing, Performance Profiling, Git, CMake
Graphics & Visualization
OpenGL, DirectX9/11, GLSL/HLSL, Signed Distance Functions, LOD, Occlusion Culling, GPU Acceleration, Computational Geometry
Domains & Tools
CAD/EDA, Signal Integrity, Physics Simulation, SDK Design (Python/Matlab/Altium), Lua Scripting, RTX 4090/5090, CUDA, Multi-GPU
| Areas of Expertise |
AI Agents & LLM Orchestration, Function Calling, Custom HTTP/TCP (C++), JSON Parsing & Validation, Context/Token Optimization, Local Inference (Ollama/vLLM), Privacy-First Architecture, C++ (11/14/17/20), OpenGL, DirectX, Shader Programming (GLSL/HLSL), 3D/2D Visualization, Signed Distance Functions, Distributed Computing, Cross-Platform Development (Windows/Linux), Performance Optimization, SDK Design (Python/Matlab/Altium), Scripting Systems (Lua), CAD/EDA Development, Signal Integrity Analysis, Physics Simulation, Computational Geometry, Multi-GPU Orchestration, VRAM Optimization |
| Environments and Tools |
C++, Python, Lua, Git, CMake, Visual Studio, GDB, OpenGL, DirectX, HLSL, GLSL, JSON/XML, HTTP/TCP, LLM APIs, Ollama, vLLM, 3ds Max SDK, SQL, SVN, CVS, Boost, STL, ARM Assembly, x86 Assembly |
| Operating Systems |
MS Windows, macOS, Linux |
Education
Master of Science, Applied Mathematics & System Programming
Voronezh State University, Russia | 1999 – 2004
Faculty: Applied Mathematics, Informatics and Mechanics | Specialization: System Programming, Computational Mathematics
Currently Conducting Projects
Simbeor / Simberian Inc.
Challenge: Engineers needed natural language control over complex CAD workflows, but existing LLM integrations were either cloud-dependent (privacy risk) or high-level Python wrappers (performance bottleneck).
✓Native C++ AI Infrastructure: Custom HTTP client in C++ for LLM communication — no external dependencies, full control over streaming, headers, error recovery; robust JSON parser for tool_calls validation; context compression layer reducing token payload by 30–50%
✓Agentic Workflow Engine: Function-calling framework with physics-grounded validation; autonomous post-layout analysis (net selection → SICA/ERC → violation extraction → report generation); multi-provider abstraction (OpenAI, Ollama, Qwen, OpenRouter)
✓Privacy-First Local Deployment: Local LLM inference on dual RTX 4090+5090 (56GB VRAM); model parallelism + context sharding for large engineering contexts; quantization support (GGUF/AWQ) for efficient multi-GPU inference
Result: Autonomous AI orchestrator for electromagnetic CAD with native C++ implementation, supporting 4+ LLM providers, 30–50% token savings, and full on-premise deployment capability.
Simbeor / Simberian Inc.
Challenge: PCB designs contain millions of geometric elements; rendering them interactively with impedance visualization, pin highlighting, and semi-transparent layers required breakthrough optimization.
✓High-Performance OpenGL Pipeline: Signed Distance Functions (SDF) for semi-transparent layer rendering; dynamic LOD + occlusion culling maintaining 60 FPS with million-element PCB assemblies; GPU-accelerated field visualization
✓Engineering-Grade Visualization: Pin/impedance overlay with adaptive labeling; chiplet support with hierarchical visibility controls; cross-platform consistency (Windows/Linux) with OpenGL fallback
✓Integration & Extensibility: Unified UX with Board Analyzer & Compliance Browser; performance metrics collection system; SDK exposure for Python/Matlab programmatic control
Result: Interactive 3D/2D visualization for complex PCB designs with engineering-grade accuracy.
Recently Completed Projects
Simbeor / Simberian Inc.
- Designed "thin client" agent architecture with job distribution protocol, priority queues, retry logic, and compression
- Implemented load balancing across multiple agents with dynamic assignment based on CPU/GPU availability
- Achieved full functional parity between Windows/Linux agents for enterprise heterogeneous environments
- Built remote crash diagnostics: automatic stack trace capture, symbol resolution, and report upload
- Developed Lua scripting engine for QA automation and customer-defined workflows without recompilation
- Exposed distributed execution via Simbeor SDK (Python/Matlab/Altium) with documented API and examples
Result: Scalable distributed computing infrastructure supporting enterprise deployments, with Linux parity and production-grade observability.
Simbeor / Simberian Inc.
- Designed unified C++ core with language bindings (Python via pybind11, Matlab via MEX, Altium via COM)
- Implemented JSON-based data exchange for consistent serialization of stackup, geometry, results across languages
- Added thread-safe API enabling parallel simulation submission from external orchestrators
- Developed functions for stackup import/export (EPX, XML, JSON, XLSX) enabling AI agents to manipulate design data
- Provided comprehensive documentation with runnable examples, reducing onboarding time from days to hours
Result: Multi-language SDK ecosystem enabling AI-driven workflows, adopted by partners for automated compliance checking and design optimization.
Achievements
| ✓ | Native C++ AI Orchestrator for CAD — Simbeor Commander with custom HTTP/JSON, function-calling, physics-grounded validation |
| ✓ | 30–50% Token Reduction — intelligent context compression for engineering data, lowering LLM cost and latency |
| ✓ | Privacy-First AI Deployment — local inference on RTX 4090+5090, zero cloud dependency for sensitive IP |
| ✓ | Interactive 3D/2D for Million-Element PCBs — SDF-based rendering, dynamic LOD, GPU acceleration maintaining 60 FPS |
| ✓ | Cross-Platform Distributed Computing — "thin client" architecture with Linux parity, load balancing, remote diagnostics |
| ✓ | Multi-Language SDK Ecosystem — Python/Matlab/Altium bindings enabling AI/ML integration and workflow automation |
Hardware & Infrastructure
Local AI Development Setup:
NVIDIA RTX 5090 32GB
NVIDIA RTX 4090 24GB
56GB Combined VRAM
128GB RAM
Local LLM inference (Ollama, Qwen, Llama 3) with quantization; model parallelism and context sharding for large engineering contexts; no cloud dependency for sensitive CAD/IP data.