Digital Workforce
A multi-tenant platform to build, deploy, and manage AI agents across chat, web, and voice, grounded in RAG over each tenant's knowledge base.
Architecture at a glance
data flows left to right- Chat / Web / Voice
- Agent runtime
- RAG retrieval
- Vertex AI
- Tenant knowledge
connectedGemini LiveTwilioStripe
The problem
Businesses want AI agents that answer from their own documents, on whatever channel the customer uses. Building a separate bot for chat, the website, and the phone line does not scale.
What I built
A multi-tenant platform where each business builds, deploys, and manages its own AI agents. An agent is defined once and answers on chat, on the web, and over voice calls.
Engineering decisions
- Grounded answers. Responses are built on retrieval over the tenant’s own knowledge base, not on the model’s general knowledge.
- Tenant isolation in retrieval. Each tenant’s knowledge stays scoped to that tenant’s agents.
- Voice in real time. Phone conversations run over Twilio with Gemini Live for low-latency speech.
- Billing built in. Stripe billing is part of the same platform.