Выспрашивал про то как и чо мы там делали с MCP Unreal Engine, потом предложил, что это можно добвить в резюме :) Типа ахуеть уникальный опыт, ни у кого такого нет (тут я не согласен, но спорить не буду). Как по мне просто очередной инженерный велосипед. (воспринимаем пост как юмор)

Project: Deterministic LLM-to-Engine Automation Pipeline for Unreal Engine Material Graphs via MCP Challenge: Automating the generation of complex Unreal Engine Material Graphs using LLMs was severely bottlenecked by a complete lack of official documentation for Unreal MCP, critical semantic name collisions between the protocol and the engine's internal C++ API, and the probabilistic nature of LLMs, which frequently caused context drift, invalid pin typing, and cyclic graph dependencies. Key Contributions: Reverse Engineering & Introspection: Designed a dynamic service discovery layer. Developed scripts that queried the active Unreal MCP API at runtime to fetch live data schemas, node definitions, and pin configurations, ensuring seamless forward compatibility with engine updates. Multi-Level Intermediate Representation (MLIR): Implemented a cascaded pipeline driven by 5 isolated context contracts (Harness prompts). The architecture enforces a gradual reduction of abstraction: translating high-level material concepts into a cycle-free logical topology, and then compiling it into a strict JSON payload. State Management & Static Analysis: Utilized a Git-based versioning pipeline to store immutable state snapshots for each translation stage, preventing context accumulation bugs. Built a two-pass semantic linter that validated graph connectivity and strict pin types against the live API schema before compilation. Result: Achieved a reliable, production-grade automation infrastructure that guarantees 100% first-shot accuracy in building complex Material Graphs. Successfully transformed volatile, generic LLM outputs into deterministic, execution-ready assets for Unreal Engine without relying on costly closed-source AI agents.