Jeeves — B2B AI Assistants Platform

Enterprise clients needed a way to deploy customizable AI assistants that understand their specific domain — products, pricing, policies — without building from scratch. Each client wanted their own branded concierge with access to company documents and CRM data.

Built a multi-tenant Django 5 platform with branch → specialization → client hierarchy. Each client gets their own RAG knowledge base (pgvector + Qdrant), configurable LLM provider (OpenAI/Anthropic/Ollama), and a dual-agent pipeline: Jeeves handles conversations while Concierge manages escalations.

Integrated 9 MCP tool servers — RAG retrieval, email, lead management, sales intelligence, coaching, memory, escalation, XLSX processing — all composable per-client. React 18 dashboard with 7-language UI, Stripe billing, and embeddable web chat widget.

Platform serves multiple B2B clients across DACH region. Each deployment takes under 1 hour from document upload to live AI assistant. RAG accuracy improved to 92% with hybrid pgvector + Qdrant search. Clients report 60% reduction in support ticket volume and 40+ hours saved weekly on routine queries.