SourceVoice
An English and Mandarin voice interface for exploring injection molding costs, supplier questions, and negotiation prompts.
ElevenLabs Hackathon · 11 December 2025
Problem and Context
Injection molding procurement involves technical terms, cost assumptions, and supplier conversations across languages. SourceVoice explored this workflow in English and Mandarin during the four hour ElevenLabs Hackathon on 11 December 2025.
User Flow
- Describe a component, manufacturing question, and procurement context.
- Speak through browser recognition or use the text fallback.
- Review the transcript before sending technical terms or quantities.
- Inspect generated manufacturing considerations, cost categories, and negotiation prompts.
- Request ElevenLabs speech output or a generated visual reference and continue in either language.
Capabilities
The prototype covers recording, transcript review, text input, bilingual responses, cost categories, negotiation prompts, generated images, speech playback, retries, and service fallbacks.
Architecture
A Next.js and React client uses TypeScript, Tailwind CSS, and Zustand for conversation, voice, negotiation, and visualization state. Browser speech recognition produces transcripts. Next.js route handlers keep provider credentials outside the browser and call Claude, Gemini language and image services, and ElevenLabs for speech output.
Status
The public archive contains architecture, API design, feature notes, and the build timeline for the demonstration prototype.
Project Credits
I built SourceVoice during the ElevenLabs hackathon, starting from the ai-ng/swift Vercel template. The demonstration simulates an English speaking customer negotiating with an AI manufacturer responding in Mandarin.