Manuel Porras

Multi-Agent AI Platform

Serving 30+ custom AI assistants from a single runtime

The Problem

A consultancy needed to offer personalized AI assistants to each of their clients, but deploying and maintaining separate instances for 30+ configurations was impractical. Each client required unique system prompts, knowledge bases, conversation styles, and branding. The cost and complexity of running individual deployments would have made the project unviable.

The Approach

I designed a configuration-driven architecture where each AI agent is defined by a JSON file containing its system prompt, model preferences, temperature settings, and capabilities. A single Node.js runtime serves all agents, routing requests based on a shareable link identifier.

The platform supports RAG (Retrieval-Augmented Generation) with per-agent knowledge bases using chunked document embeddings. Responses stream in real-time. Conversation history persists per session, and agents can be updated by editing their configuration file without any code changes or redeployment.

A web UI provides the chat interface, while an admin dashboard shows usage analytics, session depths, and engagement metrics across all agents.

The Outcome
30+
Active client configurations
RAG
Knowledge base integration
Real-time
Streaming responses
Tech Stack
Node.jsTypeScriptOpenAI APIMySQLREST APIRAGExpress
Related

← Back to all projects