AI agents that make and answer sales calls
A sales platform with AI agents that talk to leads by phone and on the web, in real time. Teams build their own agents with no code. Agents speak English, Bangla or Japanese.
- Client
- Verbex
- Industry
- Technology and Startups
- Services
- AI and Machine Learning
The problem
Verbex wanted to build an AI SDR (sales development rep). This is software that calls leads, books meetings and answers questions the way a person would.
The AI had to talk and listen in real time, by phone and on the web. It had to speak English, Bangla and Japanese. Sales teams with no coding skills had to be able to set up their own agents.
Verbex also needed billing, payments from more than one country, and a secure account for each company.
What we built
We built the Verbex website and the console where teams create, test and run their AI agents.
- A drag-and-drop agent builder, made with React Flow, so anyone can plan how an agent talks
- Live voice with LiveKit, and Deepgram to turn speech into text while the caller talks
- Live calls in the browser, so a team can test an agent before it talks to real leads
- A knowledge base: teams add their own information, and the agent uses it to answer questions
- Call logs with a transcript and an analysis after each call
- Billing and subscriptions, with Stripe for global payments and SSLCommerz for clients in Bangladesh
- Company accounts with user roles and secure login through Auth0
- The Verbex API, so other tools such as a CRM (customer relationship management system) can connect
The result
Verbex now has a platform that sales teams run on their own. A team builds an agent without code, tests it in the browser, then uses it on real calls.
Verbex offers agents for lead generation, appointment booking, customer support, surveys, debt collection and online banking help.
Challenges we solved
- Challenge
- The agent had to talk in real time and notice when the caller interrupts.
- How we solved it
- We used LiveKit for live audio and Deepgram to turn speech into text.
- Challenge
- People without technical skills had to build agents.
- How we solved it
- We built a drag-and-drop builder with React Flow.
- Challenge
- Agents had to speak more than one language.
- How we solved it
- We set up the LLMs to work in English, Bangla and Japanese.
- Challenge
- Each company needed its own secure accounts.
- How we solved it
- We used Auth0 and built login rules that know which company a user belongs to.
- Challenge
- Clients pay from different countries.
- How we solved it
- We added Stripe for global payments and SSLCommerz for clients in Bangladesh.
