Callvo AI

Overview

Callvo AI is an outbound sales and marketing platform built around AI voice agents. A business uploads its contacts, describes how its bot should sound and what it should say, schedules a campaign — and the platform places the calls, records them, and reports back with recordings, transcripts, and summaries. I led the team that built it.

The premise is that outbound calling is a volume problem with a quality ceiling: humans do it well but don’t scale, and autodialers scale but perform badly. An LLM-driven voice agent sits in between, and the interesting engineering is everything around the call rather than the call itself.

Key Features

• Campaign creation and scheduling: Build a campaign, upload a contact list, get it approved, then schedule it — the platform runs it automatically at the appointed time.
• Bot preferences and brand core: Configurable agent behaviour and a brand definition the AI draws on, so the voice on the phone is consistent with the business it represents.
• Live agent preview: Talk to your AI assistant before a campaign goes out, to refine the approach against a real conversation instead of a script document.
• Call logs with recordings and summaries: Every call captured with its recording, transcript, and AI-generated summary, browsable per contact and per campaign.
• Lead scoring and engagement analytics: Call outcomes fed into lead scoring and engagement reporting, so the output is a ranked pipeline rather than a call archive.
• Admin approval layer: Campaigns pass through administrative review before dialling — a deliberate control given the platform makes outbound calls on a customer’s behalf.
• Full SaaS surface: Email-verified registration, password reset, subscription billing, contact management with CSV import, a content calendar, and post scheduling.

Technologies Used

• Next.js 15 + TypeScript: App Router with grouped dashboard, auth, and server route segments.
• Vapi AI: Voice agent runtime handling real-time telephony conversations.
• Prisma + NextAuth: Data layer and authentication with email verification and OTP flows.
• Stripe: Subscription billing and payment lifecycle.
• TanStack Query + Table: Server-state management and the data-dense log and contact views.
• Radix UI, Tailwind CSS, Framer Motion: Interface layer.
• Papa Parse, Vercel Blob, Nodemailer: CSV contact ingestion, recording storage, transactional email.

Challenges and Learnings

Voice agents fail differently from chat. A text assistant that stalls for two seconds is fine; on a phone call, that silence ends the conversation. Latency stops being a performance metric and becomes a correctness one, which changes how you architect the path between telephony, model, and business data.

The other lesson was organisational. Leading the team here meant most of my leverage came from decisions made before anyone wrote code — how campaigns, contacts, and call outcomes related to each other. Getting that model right early is what let features like lead scoring and engagement analytics be additions rather than rewrites.

Outcome

Callvo turned outbound calling into something a small business can actually operate: define the agent once, upload a list, and get back scored leads with recordings and summaries attached. The work fed directly into the AI-powered outbound sales platform I later drove at BXTrack Solutions, where the same approach produced a ~40% lift in sales efficiency.