Skip to content
All projects

Company project

AI Customer Support Agent

An intelligent customer support system capable of communicating with website visitors, answering business-specific questions and capturing potential customer information.

  • Conversational AI
  • Company-specific information
  • Lead capture
  • Chat interface
  • Voice interaction
  • Backend integrations

The problem

Website visitors often have simple, repetitive questions — opening hours, services, how to book — and they expect an answer straight away. When those questions wait in an inbox, interest fades and potential customers leave without sharing their details.

For small teams, answering the same questions by hand takes time away from the work that genuinely needs a person.

Our solution

We built a conversational support agent that sits on a business website, answers questions using information about that specific business, and captures the details of visitors who want to take the next step.

Each business is described through a configuration profile — its services, key facts, tone and the details it needs to collect — so the same system can be adapted to a different business without rewriting the agent.

How it works

  1. 01

    The visitor asks a question

    Through text chat or by voice, directly on the website.

  2. 02

    The agent uses business context

    The language model answers from the business profile rather than general knowledge, keeping replies relevant and on-topic.

  3. 03

    Details are captured as structured data

    When a visitor wants to book or be contacted, the agent calls a defined tool that records their details in a consistent format.

  4. 04

    The team follows up

    Captured requests appear in an admin view, tagged by business, ready for a person to act on.

Architecture & workflow

  1. Website visitor
  2. Chat / voice interface
  3. API layer
  4. LLM with tool calling
  5. Business profile
  6. Lead & booking records
  7. Admin view

Features

  • Conversational chat widget

    A clean chat interface that fits into an existing website.

  • Voice interaction

    Visitors can speak their question and hear the reply.

  • Business-specific knowledge

    Answers are grounded in each business's own information.

  • Structured lead capture

    Tool calling turns conversations into consistent, usable records.

  • Configurable per business

    New businesses are added through configuration, not new code.

  • Admin view

    Captured requests are listed and filterable for the team.

Technology stack

Interface
Next.jsReactTypeScriptTailwind CSS
AI
LLM API (Llama 3.3 70B via Groq)Tool / function callingText-to-speechBrowser speech recognition
Backend
Next.js API routesStructured data storage

Screenshots & demo

Screenshots and a demo walkthrough will be added here.

Want to see it sooner? Ask us for a live demo.

Challenges

  • Staying on topic

    A general-purpose language model will happily answer anything. The agent needed clear instructions and business context so that it stays focused and avoids guessing.

  • Reliable data capture

    Free-form conversation has to end in structured, usable records. Using tool calling rather than parsing text made captured details consistent.

  • One system, many businesses

    Keeping business-specific details in configuration rather than code made it practical to adapt the agent to different kinds of businesses.

What we learned

  • Grounding an assistant in a specific business's information matters more than the size of the model.
  • Structured tool calls are far more dependable than extracting details from free text.
  • A good agent knows the limits of what it should answer — and leaves the rest to people.

Want something similar? Talk to us.

Tell us about the process you want to improve, and we'll help work out what makes sense.

Start a Project