B2B SaaS · 4 weeks
An AI support agent that drafts, triages, and knows when to escalate.
A growing SaaS team was drowning in support tickets. We built an AI agent that reads each ticket, classifies it, drafts a reply from their docs, and escalates anything sensitive to a human.
AI agent
The challenge
Support volume was outgrowing the team. Replies were slow and inconsistent, and the best engineers were spending hours on repetitive questions instead of the product.
What we built
- An agent that reads inbound tickets, classifies intent, and drafts a reply grounded in the team's own documentation (RAG)
- Human-in-the-loop approval on every customer-facing message
- Automatic escalation and routing for refunds, security, and edge cases
- Evals in CI so prompt and model changes are measured, not guessed
The outcome
Routine tickets get a fast, on-brand draft ready for a human to approve, while sensitive issues route straight to a person — freeing the team to focus on the product.
AnthropicRAG / pgvectorPythonTypeScript
Want something like this?
Bring a brief, a sketch, or just an idea — you'll leave with a scope and a timeline.