Echo - An AI powered Support Assistant

Thumbnail
Operator Dashboard
Knowledge Base
Widget Customization
Integrations
Vapi Plugin
Billing
Widget Integration in HTML file

Inspiration

Every website today struggles with one big challenge: slow, inefficient, and expensive customer support. Users wait too long. Support teams get overwhelmed. And AI chatbots alone often fail to understand real problems. I wanted to build something that feels as smooth as Intercom, as smart as a modern AI model, and as reliable as a human agent — all in one unified system. That became Echo.

What it does

Echo is a smart support assistant designed to make customer help feel faster, smoother, and far less frustrating — for both users and support teams. Instead of juggling separate tools for chat, voice calls, and ticket escalation, Echo brings everything together in one clean, simple widget that any website can use.

You can visit the operator dashboard at 👉 link Then you can try embedding it in a html file.

Echo focuses on one goal: help people get the right support anytime.

Features:-

Smart Knowledge Base using Embeddings and RAG: The admin can store files related to the questions that can be asked in the chat. The AI uses these files data to answer the question.
Human Handoff & Auto-Close using AI Tools: If the answer to the question is not found in the knowledge base or the user shows frustration then the conversation is escalated to a human operator. If the conversation is ended then the AI auto-closes the chat by marking it as resolved.
Voice Support by VAPI: The AI voice call feature is powered by Vapi. The users can voice call the AI and chat or call on the given phone number on which the AI will respond to!
Authentication & Billing by Clerk: Authentication is securely powered by Clerk. Some features like the knowledge base, team invites, VAPI Integration & AI Features are paid. Clerk billing is used for secure payments.
Widget Embedding: The operator can visit the integrations page and copy the code and add it to their website which adds a small chat icon on the corner. This widget can be used by the users to chat and make voice calls.
Human Take-Over: Whenever the human operator feels the AI is going in the wrong direction the human operator can interrupt and take over the conversation.
User Details: On the admin dashboard, the user's details are shown like the name, country, session details, browser and OS of the user for easier fix to the problem.
Widget Customization: The widget appearance can be edited by the admin of the organization in this section. The options like the first message, suggestions etc... can be edited here.

How we built it

Tech Stack

Frontend: Next.js 15, TailwindCSS, Shadcn/UI
Backend: Node.JS, Convex
Secrets Manager: AWS
AI Chat Features: OpenAI, Convex AI
Authentication & Billing: Clerk
Voice AI: Vapi
Deployment: Vercel

Architecture Overview

Built a custom support widget with React + iframe embedding
Implemented chat AI pipeline with routing + session context
Integrated Vapi voice agents for real-time voice conversations. We have used the method of white-labeling API keys. So the user gets their API key from Vapi and then enters that key into our website which then connects to Vapi.
Created a human escalation system that triggers when:
- AI can’t solve a query
- The workflow detects a support failure
- User explicitly requests “talk to a human”
Added operator dashboard for handling escalations
Streamed messages + voice events in real-time

Challenges we ran into

Syncing voice and chat contexts so the AI never “forgets” past messages
Building a clean escalation flow without breaking the chat session
Handling iframe communication between widget and main app
White labelling API keys to AWS Secrets Manager
Rate limits + reliability issues while testing multiple AI providers
Maintaining low-latency voice responses
Balancing AI accuracy with user experience

Accomplishments that we're proud of

Developed a fully functional AI support assistant end-to-end
Integrated live voice support
Implemented smooth human escalation that feels natural
Designed a clean, embed-anywhere widget
Built a system that’s practical and usable, not just a prototype
Achieved a near-real support experience similar to enterprise tools like Intercom