AI

AI FAQ Assistant

An AI assistant that answers customer questions from the company’s own FAQ knowledge base and sends anything urgent straight to the team.

Client
US home-services brand
Type
AI
Scope
HubSpot workflows · HubDB · Embeddings · Serverless functions
AI US home-services brand
  1. 01 Customer question
  2. 02 Embed & search FAQs
  3. 03 Blended re-ranking
  4. 04 Urgency check
  5. 05 Answer or hand-off

The project

About the project

Customers ask many of the same questions again and again. The goal was an assistant that answers from the company’s approved FAQ content, never invents answers, and knows when a message needs a person.

The assistant runs inside HubSpot workflows. FAQs live in a HubDB table. Each incoming question is matched against them with embeddings, then re-ranked with a blended score of semantic similarity and keyword overlap, so short or oddly worded questions still find the right answer. An urgency taxonomy flags time-sensitive messages and hands them to the team instead of replying automatically. The first version was later redesigned into a single-function architecture that is simpler to maintain and cheaper to run.

Key features

Private client work. The client’s name is kept confidential.

  • Answers only from the company’s own HubDB FAQ table
  • Embedding search with blended semantic and keyword re-ranking
  • Embedding cache to keep responses fast and costs low
  • Urgency detection that routes time-sensitive messages to people
  • Single-function architecture inside HubSpot workflows

Contact · usually replies in 0–4 hours

Send me your
Figma file.

Or your current website. I’ll tell you the best approach: custom theme, rebuild, migration, or just the fixes that matter.

Tell me about the project

    Thanks, talk soon.

    Your message is in my inbox. I usually reply within a few hours.