Dmitry Karataev
Software Engineer
8+ years of C#/.NET backend engineering. Building open-source MCP servers that give AI assistants access to compilers, debuggers, and build systems — bridging the gap between AI and real developer tooling.
What I Do
Backend engineering by day, agent infrastructure by passion.
Backend & Architecture
Microservices, REST API, GraphQL, SOA. Full SDLC experience: from monolith decomposition to DI adoption, unit testing, and Agile process design.
Agent Infrastructure (MCP)
A suite of Model Context Protocol servers giving AI assistants access to Roslyn, DAP debugger, build/test systems, and multimodal inputs. 1.5+ years of daily AI-assisted engineering.
agent-first-learn
A practical methodology and open guide for productive human–AI collaboration: context discipline, guardrails, knowledge management, and verifiable results.
Open Source Projects
MCP servers — giving AI assistants real developer tools, not just file text. Why these projects?
Desktop IDE for human–agent work on .NET: cockpit-style attention (PFD/MFD), in-proc MCP, Roslyn/debug/build integration, environment readiness (Dark Cockpit).
Agent memory and knowledge-base MCP 2.0: one TOML config (SSOT with the IDE), hot-context, routing, localhost status on loopback. Public KB: kb-public. Why one config?
Public read-only knowledge-base slice (CC BY-SA): playbooks, integrity kernel, routed domains — material for agents without the full private canon. Pairs with agent-notes-mcp.
MCP over Roslyn: diagnostics, quick fixes, go-to-definition, find usages, rename, solution structure. AI gets the same code analysis layer as the IDE.
.NET debugging via DAP (netcoredbg): breakpoints, launch/attach, stack traces, variables, stepping, continue/stop. Reproducible debugging from chat.
dotnet build / dotnet test with single-flight queue and structured error output. AI sees the essence of failures, not log walls.
Webcam & screen capture, audio, burst sequences (capture + analysis), Whisper transcription. Multimodal input for the assistant pipeline.
Hybrid full-text (and optional semantic) index over a workspace: MCP search, explain, verify identifiers — grounded retrieval instead of guessing symbols.
Git operations via MCP: status, diff, logical commits, preflight — the same repo hygiene an agent needs beside the compiler.
A practical guide to productive AI-assisted engineering: context management, summarization protocols, guardrails, energy-first training, and the environment-over-model philosophy.
Experience
8+ years of backend development across energy, document management, and enterprise systems.
Engineering data management systems for EPC digital twins. .NET 9, ASP.NET Core, Blazor WASM, GraphQL (HotChocolate), SQL Server. Introduced DI, unit testing, Agile practices, and a team knowledge base.
Cross-functional SDLC: analysis, documentation, development, estimation. .NET backend systems.
.NET Core web services and SOA controllers. Entity Framework Core, REST API, SQL Server stored procedures.
Electronic document storage on IBM ECM. TWAIN scanning pipeline, SAP integration, microservice extraction from monolith. Memory dump debugging, .NET Reflection, COM interop, Rx.NET.