Building VoiceCake: A Multi-Channel AI Voice and Chat Platform

Problem

Businesses needed a way to launch human-sounding voice and chat agents that could support, sell, and automate across phone, web, WhatsApp, SMS, and email — without stitching together separate providers, admin tooling, and compliance infrastructure themselves.

VoiceCake needed a platform that could handle that end to end: agent management, voice provider integration, embeddable deployment on client sites, QA/compliance visibility, CRM syncing, and a way for enterprise developers to actually build on top of it.

Solution

I worked as a full stack developer on VoiceCake.io, building core platform infrastructure end to end.

Admin portal and user-facing UI

I built the admin portal from the ground up and fully revamped the user-facing UI — giving operators and end users a cohesive product surface instead of a patchwork of internal tools.

Multi-provider voice stack

I integrated multiple voice and telephony providers — LiveKit, Twilio, Deepgram, Hume, Cartesia, and Hamsa — so agents could run on different underlying voice engines depending on the use case.

I also built voice cloning and inference testing tooling so agents could be validated before go-live rather than discovering issues in production.

Embeddable web widget

I built an embeddable web widget so businesses could deploy an agent directly on their own site without custom integration work on their end.

QA, compliance, and CRM sync

For QA and compliance, I built call logs with full transcripts, giving teams a complete record of every agent conversation.

I built CRM integrations to sync outcomes, leads, and follow-ups back to systems of record — so agent conversations translated into actual pipeline data instead of living in isolation.

Developer API and evaluation workflows

I built out API docs and API infrastructure for custom enterprise integrations, plus a developer-facing API playground for testing against the API directly.

I also worked on AI evaluation workflows to assess agent performance before and after deployment.

Tech stack

Frontend: Next.js, React

Backend: FastAPI, LangChain, MCP

Voice / telephony: LiveKit, Twilio, Deepgram, Hume, Cartesia, Hamsa

Infrastructure: AWS (S3, DynamoDB), Vercel

What I would tell teams evaluating voice AI platforms

1. Prioritize latency and voice quality — users forgive slow chat, not slow phone agents.
2. Build eval workflows before scaling — bad agents at volume are worse than no agents.
3. Plan for CRM integration on day one, not as a phase-two afterthought.
4. Give operations teams call logs and transcripts, not just developer APIs.
5. Ship embeddable deployment early — most clients want agents on their site without a custom integration project.

---

Hire / next step: Voice AI Developer · Full Stack AI Developer · Voice AI services · Book a call