An AI support agent for TaskFlow, a project management SaaS, that answers documented questions instantly and hands off to a human the moment it isn't sure — no guessing, no invented policy.
Most AI chatbots will confidently invent an answer rather than admit they don't know — which erodes trust faster than a slow human reply ever would. This system is built the opposite way: it only answers from TaskFlow's real documentation, and the moment a question touches billing, security, or anything it can't verify, it says so and routes the customer to a real person.
One accurate answer, one honest escalation, and the record landing in Airtable — unscripted, recorded live.
This is the live agent, connected to the real knowledge base. Ask it anything a TaskFlow customer might ask.
One question, four possible outcomes — all handled without a human touching it, unless a human is actually needed.
The customer asks anything — a how-to, a billing question, a bug report, or something unrelated entirely.
Dify searches TaskFlow's help documentation using hybrid semantic + keyword search, ranked by relevance.
If the docs clearly cover it, the AI answers. If it's billing, security, a bug, or unclear — it escalates instead of guessing.
Escalations are logged straight into Airtable through a simple handoff form, ready for the support team to follow up.
Knowledge base, RAG retrieval, and the conversational AI layer
text-embedding-3-small for semantic search, GPT for responses
Escalation log and the customer-facing handoff form