Scaling AI Knowledge Systems: Lessons from the DHTMLX MCP Server

2026-03-19 · Source: Artificial Intelligence in Plain English - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, medium

Summary

DHTMLX developed the MCP Server, a centralized knowledge layer that provides AI assistants and developer tools with structured access to current documentation across its product line, including Suite widgets, Gantt, and Scheduler. This system utilizes a Retrieval-Augmented Generation (RAG) approach, which retrieves relevant documentation fragments and feeds them to Large Language Models (LLMs) to generate context-grounded responses. While effective for single products, scaling RAG across multiple distinct products, each with unique documentation, introduced complexity. Initial attempts to use a single vector index led to accuracy issues due to mixed contexts. Separating knowledge into product-specific indexes improved accuracy but necessitated a fast and flexible query routing mechanism. DHTMLX addressed this by developing a custom machine learning model for domain classification, optimized for low latency and high accuracy through distillation and 8-bit quantization, achieving performance comparable to TinyBERT while maintaining a smaller footprint.

Key takeaway

For AI Engineers building RAG systems that span multiple distinct product lines, relying on a single, unified knowledge index will likely degrade answer quality. You should instead segment your knowledge base by product and implement a lightweight, specialized machine learning model for query routing. This approach, leveraging techniques like distillation and quantization, ensures both accuracy and the low latency required for real-time AI assistance, preventing context mixing and improving developer efficiency.

Key insights

Scaling RAG systems for multiple distinct products requires intelligent query routing and optimized knowledge structuring.

Principles

• Separate knowledge bases for distinct product domains.
• Balance model size and accuracy for real-time routing.

Method

A custom machine learning model for domain classification was developed using distillation and 8-bit quantization to achieve high accuracy and low latency for routing queries to product-specific RAG indexes.

In practice

• Implement product-specific vector indexes.
• Use distillation for smaller, capable models.
• Apply 8-bit quantization to reduce model size.

Topics

• Retrieval-Augmented Generation
• Model Context Protocol
• Machine Learning Models
• Model Distillation
• Quantization

Best for: AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• Grounding Your LLM: A Practical Guide to RAG for Enterprise Knowledge Bases — RAG systems combine LLMs with dynamic knowledge retrieval to provide accurate, auditable, and updatable answers for enterprise data.
• How RAG, GraphRAG, and Context Engineering Improve AI Performance — Effective AI performance hinges on context engineering, not just raw model intelligence.
• Knowledge Layers: A Practical Fix for Stale RAG in Agentic AI — Pre-building policy-enforced knowledge layers before agent execution resolves stale RAG issues in multi-step AI agents.
• Retrieval Augmented (Knowledge Graph), and Large Language Model-Driven Design Structure Matrix (DSM) Generation of Cyber-Physical Systems — LLMs, especially with RAG and GraphRAG, can automate Design Structure Matrix generation for complex systems.
• Reducing Hallucinations in Complex Question Answering using Simple Graph-based Retrieval-Augmented Generation (long version) — Integrating lightweight knowledge graphs into RAG systems significantly reduces LLM hallucination and improves complex QA accuracy.
• Don't Retrieve, Navigate: Distilling Enterprise Knowledge into Navigable Agent Skills for QA and RAG — Corpus2Skill distills document corpora into navigable hierarchical skills, enabling LLM agents to actively explore and combine evidence.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence in Plain English - Medium.