Each one with the work that shows it.
- AI agents and MCP
- MCP servers with their own OAuth 2.1, LLM tool calling, agents that run on several models
- Shown in Team OS, OrbiBot's agent
- AI safety
- Human approval for money actions, confirmations the model can't skip, every AI action under the user's own permissions
- Shown in Team OS
- LLM engineering
- Structured output against a schema, grounding in a database, server-side checks, model rotation, prompts tested before and after each change
- Shown in RoboAI Shop's parts planner
- RAG
- Textbook transcription, chunking, pgvector embeddings, answers that cite the page
- Shown in aiTutor's physics questions and answers, over a 366-page textbook
- AI-assisted development
- Claude Code with skills, subagents and headless runs; AI review agents
- Shown in RoboAI's repositories
- Backend
- C#, .NET 10, ASP.NET Core, CQRS, RabbitMQ, SQL Server, EF Core, Redis, MongoDB, Azure
- Shown in Orbitax
- Web
- TypeScript, Next.js, React, Angular, Supabase with row-level security, Tailwind
- Shown in Team OS, RoboAI Shop, OrbiBot's dashboard
- Robotics
- ROS 2 Jazzy, Nav2, slam_toolbox, robot_localization, URDF, RViz, MuJoCo, Gazebo
- Shown in OrbiBot
- Embedded and electronics
- STM32 firmware, PID control, ESP32, Arduino, Raspberry Pi, KiCad
- Shown in OrbiBot V1's firmware, the earlier robots
- Computer vision
- OpenCV, MediaPipe, PyTorch, YOLOv8
- Shown in Robot Niko, EnviroBot, sign-language detection
- Mechanical
- Welding, design for 3D printing, FreeCAD scripted in Python
- Shown in OrbiBot V1's frame, Robot Niko
- Problem solving
- Codeforces Specialist, best rating 1509, more than 1,200 problems solved
- Shown in Codeforces