How does live agent handoff work?

The backend changes the conversation state, notifies an available agent and keeps the existing chat history attached to the same session.

Can the widget be added to an existing website?

Yes. An embeddable widget can connect to the support backend without replacing the main website or application.

How does the chatbot reduce unsupported answers?

It answers from retrieved approved sources and uses escalation when the available context is not strong enough for a reliable response.

Does the agent receive the customer conversation history?

Yes. The human agent joins the same conversation with the earlier messages and session context already available.

What happens to questions the chatbot cannot answer?

They enter a knowledge-gap queue that helps the support team identify missing or unclear documentation.

Automation case studyAnonymized client project

A support chatbot that hands unresolved questions to a real agent

Ampere Support combines an embeddable AI chat widget with a live agent dashboard. It answers from company knowledge, recognizes when the available information is not enough and moves the same conversation to a human agent.

In short

Ampere Support is a RAG customer support chatbot with live agent handoff. Businesses embed the chat widget on their website, where it answers from approved company documents. When it cannot answer confidently, it escalates, alerts the team on Discord and opens the same conversation in an agent dashboard, so customers do not repeat themselves and unanswered questions feed a knowledge-gap queue.

Retrieval: LangChain, Together AI
Realtime chat: Socket.IO
Escalation alerts: Discord
Handoff: Same conversation, full history

View the original project presentation (PDF)

01 / The problem

Why the RAG support chatbot was needed

A chatbot becomes frustrating when it guesses, repeats itself or ends the conversation after failing to answer. Support teams also lose time when an escalated customer has to explain the same issue again.

The system needed a reliable transition from AI to human support while keeping the conversation, customer details and relevant knowledge available to the agent.

02 / The build

How I approached the build

I built an embeddable Next.js widget and a Node.js backend using Socket.IO for realtime communication. LangChain and Together AI power retrieval from company knowledge before the assistant prepares a response.

When the system cannot answer confidently, it creates an escalation, alerts the team through Discord and opens the same conversation in the agent dashboard. Unanswered questions also feed a knowledge-gap view for future improvement.

03 / Capabilities

What the RAG support chatbot does

Embeddable chat widget

The support experience can be added to an existing website without rebuilding the site.

RAG answers

Responses use approved company documents instead of relying only on general model knowledge.

Automatic escalation

Unresolved conversations can move to the support queue based on clear conditions.

Live agent handoff

An agent joins the same chat with the earlier messages and customer context.

Realtime dashboard

Socket.IO keeps conversation and availability states synchronized.

Knowledge-gap queue

Questions the system could not answer become visible for content improvement.

04 / Workflow

How it works, step by step

01
Receive the question
The visitor starts a conversation through the embedded widget.
02
Retrieve company context
The backend searches the approved knowledge source for relevant information.
03
Answer or escalate
The assistant responds when grounded context is available and escalates when it is not.
04
Continue with an agent
A human joins the existing conversation with the full history visible.

05 / Product screens

Product screens

Select a screen to inspect the interface, workflow and operational details more closely.

Product walkthrough

01 / 06

Ampere Support — presentation, page 1

06 / What changed

The practical result

Customers can move from automated help to a person without starting a second conversation.
Agents receive the earlier messages and context before they reply.
Unanswered questions become an actionable list instead of disappearing from the support system.

Common questions

Questions about building a similar RAG support chatbot

It retrieves relevant information from approved company sources before generating an answer, which helps keep responses grounded in the business knowledge base.

Work with me

Need a RAG chatbot developer?

I build support assistants that use your knowledge, expose uncertainty and give human agents a clean path into the conversation.

Discuss your project

Keep exploring

View all projects

Enterprise

A support chatbot that hands unresolved questions to a real agent

Why the RAG support chatbot was needed

How I approached the build