Corporate Soul Starter

Multi-Agent RAG

A microservices-based multi-agent knowledge system for corporate environments. Five specialized agents (HR, Legal, IT, Finance, and a corporate-soul orchestrator) handle different domains with RAG-backed answers. Built with Kubernetes-ready architecture, S3→SQS→Textract document ingestion, and session memory.

Impact & Results

Production-ready multi-agent architecture

Automated document ingestion handling PDFs and transcripts

Scalable Kubernetes deployment

Structured markdown/table rendering for AI responses

Technical Architecture

Multi-agent orchestration with domain specialization OpenSearch vector store for RAG S3→SQS→Textract document pipeline Redis session memory Kubernetes-ready Helm charts

Challenges & Solutions

Challenge 1

Different corporate departments need different knowledge domains

Solution 1

5 specialized agents with domain-specific RAG indexes

Challenge 2

Documents arrive in mixed formats (PDFs, transcripts, scans)

Solution 2

S3→SQS→Textract pipeline for automatic PDF and transcript processing

Challenge 3

Responses must be contextual to the user's conversation history

Solution 3

Redis-backed session memory (30 turns, 72h TTL) for conversational context

Challenge 4

System needs to scale across large organizations

Solution 4

Helm/K8s deployment for horizontal scaling

Technology Stack

FastAPI OpenAI GPT OpenSearch (RAG) Redis (session memory) React Vite Helm/K8s S3 SQS AWS Textract AWS Transcribe