Menu
Theme

Years I contributed:2025AI · Voice application

Vocal.Fun

An AI voice application built from scratch, combining live conversations, wallet access, and recordings people can download and share.

My role
Project manager & software development engineer
Context
Client project · Built from scratch
Explore the project

Vocal.Fun lets people discover voice agents and talk to them through a character-led interface. I worked as both project manager and software development engineer, building the front end from scratch with two teammates.

Our scope covered the application’s front end, including voice calls, wallets, and audio sharing and downloads. Audio streaming and the runtime cost of recording made this a particularly interesting project for me, alongside the idea of speaking directly with AI characters.

Built with
NuxtVueTypeScriptPiniaWeb Audio APISocket.IOFFmpeg.wasmEthers
  • The application and call experience

    Shared front-end implementation with two teammates, connecting the character interface to live calls. My work included backend integration, microphone-permission checks, error handling, and call states.

  • Recordings that can leave the app

    Built downloads and sharing controls so people could save a conversation as a video with the agent’s portrait. Added Web Share API support and links back to the application in shared messages.

  • Wallets and audio behavior

    Added wallet connection flows, refined balance presentation, and built a reusable audio service for interface sounds and playback.

From a live audio stream to a shareable recording

A live call and a saved recording need different handling. The call captures microphone audio, sends it over Socket.IO, and plays incoming PCM audio through the Web Audio API. Exporting that conversation adds encoding work to a browser already handling audio.

I used FFmpeg.wasm for the export. Its runtime loads when an export is requested and can be reused for later downloads. The conversion takes the recorded samples, encodes the audio, and combines it with the agent’s portrait to produce an MP4.

The portrait is scaled and padded to fit the output without stretching, and the video runs for the recording’s duration. After a successful export, the implementation removes the temporary files from FFmpeg’s in-memory filesystem. People can save or share the resulting video directly from the application.

My independent practice

My independent work spans client collaborations and my own developer tools. I established my business in January 2025, building on earlier project work. I handle architecture, implementation, and delivery; when a project needs more capacity, I can bring designers and engineers through DeSource Labs.

Career timeline January 2025–present

  1. Individual Entrepreneur / Software Development Consultant

    January 2025–present

Broader responsibilities

Hands-on product development and technical consulting through my own business.

  • Develop web applications, front-end systems, and integrations for clients, from technical planning through implementation and handover.
  • Work on software architecture, backend and infrastructure planning, AI integrations, and blockchain applications.
  • Create and maintain developer tools including Phone Mask, DeSource Image, Context7 Widget, and Browser AI.
Full CV
Start a conversation

What are you
working on?

An engineering role, a product to build, or a technical problem to work through. I’d like to hear about it.