Stemma

Overview

Stemma was a Danish-speaking AI phone assistant designed to answer inbound calls for businesses that couldn’t always pick up the phone. The goal was to handle common customer interactions—such as answering questions, taking bookings, and scheduling appointments—using natural voice conversations.

Over several months, we built a working prototype, spoke with potential customers, and secured a pilot customer before ultimately deciding not to continue the company. Although the project ended, it gave me valuable experience building and validating a real-world AI product from scratch.

What I built

• Real-time voice AI pipeline. Designed and implemented an end-to-end pipeline that connected speech-to-text, an LLM, and text-to-speech to enable natural phone conversations in Danish.
• Backend & telephony integration. Built the backend services in Python/FastAPI and integrated the system with Twilio to handle real-time phone calls and conversation state.
• Conversation orchestration. Developed the application logic that managed prompts, conversation flow, and interactions between the different AI services while keeping latency low enough for live conversations.
• Rapid product iteration. Worked closely with potential customers to understand their workflows, translate feedback into product improvements, and continuously refine the prototype.

What I learned

Stemma reinforced that building AI products is about much more than choosing the right model. The real engineering challenges lie in connecting multiple systems into a reliable user experience—handling latency, edge cases, and failure modes while keeping interactions natural.

It also taught me that building a successful startup requires both product-market fit and founder-market fit. While we saw encouraging customer interest, we ultimately decided that this wasn’t the opportunity we wanted to dedicate the coming years to pursuing.

My responsibilities

• AI systems architecture
• Backend engineering
• Voice AI integration (ASR, LLMs, TTS)
• Cloud deployment
• Customer discovery
• Product strategy