The latest articles on DEV Community by Abhishek Roy (@abhishek_roy_789b16fc14d8). https://dev.to/abhishek_roy_789b16fc14d8 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4056549%2Fe16c4231-87b7-43af-b8f8-6c13551a9196.png https://dev.to/abhishek_roy_789b16fc14d8 en Abhishek Roy Fri, 31 Jul 2026 11:54:49 +0000 https://dev.to/abhishek_roy_789b16fc14d8/building-task-optimal-mixture-of-experts-2man https://dev.to/abhishek_roy_789b16fc14d8/building-task-optimal-mixture-of-experts-2man <p>My name is Abhishek Roy, based in Kishanganj, Bihar. When building local AI agents, one major problem stands out: How do you route low-complexity prompts to fast, cheap models while preserving deep multi-pass reasoning for complex engineering prompts?</p> <p>In this article, I’m walking through the design of TO-MoE, a local terminal-based Python prototype built on LangGraph.</p> <p>Key Technical Sections:</p> <p>Why Keyword Routing Fails (RFC Trap A vs Trap B)</p> <p>Semantic Coordinate Routing with Cosine Similarity</p> <p>Implementing Action Verb Dominance (AVD) &amp; 1-Token Jury</p> <p>State Protection: Silent Escalation &amp; Quarantine Gates</p> <p>Memory Architecture: ChromaDB Vector Store + Context-Assist Semantic Cache</p> <p>Benchmark Results &amp; Lessons Learned</p> ai architecture machinelearning python