Documentation Platform & AI Assistant
Automated documentation platform featuring multi-language Astro Starlight generation, RAG conversational chatbot, and universal Model Context Protocol (MCP) server, built during the 15-week internship at DevScope (Porto).
Automating technical documentation and AI reasoning.
Developed as the core outcome of the 15-week curricular internship at DevScope (Porto), this platform replaces commercial SaaS tools like GitBook by transforming Markdown repositories into multi-language static portals with zero external cloud dependencies.
The solution integrates 6 modular components: RagApi (.NET 8 for ingestion), RAG Server (Node.js inference engine), Chatbot Widget (embeddable IIFE JS widget), Backoffice (React 19 configuration portal), devscope-docs-engine (TypeScript CLI for CI/CD pipelines with block-level incremental translation), and devscope-mcp (Python Model Context Protocol server exposing knowledge to AI agents).
The Problem & Key Challenges
DevScope relied on paid third-party platforms like GitBook ($300+/yr) to publish product documentation. These solutions presented high recurring subscription costs, restricted customizability, vendor lock-in, and lack of control over how internal documentation data was processed.
Standard static site search mechanisms (like full-text keyword matching) failed when engineers asked questions in natural language using different terminology, making an intelligent RAG assistant and multi-product isolation essential.
- Strict multi-tenant data isolation across products at vector database level (Azure CosmosDB)
- Vendor-agnostic AI architecture abstracting LLM, embedding, and vector providers via configurable interfaces
- Incremental block-level translation using Azure AI Foundry models without re-translating unchanged Markdown chunks
- Universal Model Context Protocol (MCP) server exposing documentation to AI agents (Claude Desktop, Cursor, Copilot)
- Deploying static sites automatically to Azure Blob Storage ($web container) via CI/CD pipelines
Architecture & Technical Decisions
Layered modular architecture with clear separation between Markdown ingestion (devscope-docs-engine), REST ingestion API (RagApi), RAG inference engine (RAG Server), MCP server (devscope-mcp), and isolated presentation widgets.
- Complete vendor independence: LLM, embedding, and vector database providers are fully abstracted via environment variables
- CI/CD native CLI (devscope-docs-engine) executes directly within build pipelines, versioning translations inside Git
- MCP server allows any external AI agent (Cursor, Copilot, Claude) to query product knowledge uniformly
- Strict multi-tenant isolation prevents data contamination between distinct DevScope products
Tech Stack & Infrastructure
Project Diagrams & Interfaces
Generated Astro Starlight documentation portal with integrated RAG AI assistant
RAG inference flow and hallucination mitigation guardrails
Incremental documentation synchronization & translation pipeline
Conversational RAG chatbot widget interface
High-level platform architecture diagram
React configuration backoffice portal
Universal Model Context Protocol (MCP) server validation