Denise

Multi-tenant AI lead-handling platform

Architect / lead developer · Kipanga

A multi-tenant AI email-triage and lead-handling engine on n8n Cloud, designed config-as-data to scale across franchise stores. Onboarding a store is a config row, not a copied workflow.

What I built

• 01Architected and shipped a multi-tenant AI email-triage and lead-handling engine on n8n Cloud, migrating off Relevance AI and designing it config-as-data so onboarding a store is a config row, not a new workflow.
• 02Replaced a brittle chained-agent legacy design with a two-tier architecture: per-store ingress calls a secret-free shared engine through a synchronous sub-workflow.
• 03Built the triage and drafting engine on OpenAI GPT-4-class models with a deterministic pre-LLM hard-stop, a schema-validated I/O contract that fails closed, and explicit routed outcomes.
• 04Reverse-engineered an undocumented CoreBridge EVO CRM write path (a lead is a Company plus a linked, activated Contact) including a silent activation gotcha, with idempotent dedupe. Proven live end to end.
• 05Integrated Microsoft Graph for inbox polling and in-thread reply drafting, and enforced a hard safety invariant: human-escalated leads never auto-send.

Stack