This semester, our lab took a course whose term project challenged us to create a AI service for the public. My initial proposal was a Chatbot that would let students chat with protagonist of the novel they were reading in class (similar to Character.ai). We presented the idea, with a rough technical plan, to the professor.

During feedback the professor suggested a small but timely pivot: instead of a fictional character, why not let citizens talk to AI personas of the South Korean presidential candidates? The 2025 election was approaching, and such a service could increase the engagement from the users. We liked the idea, and decided to add voice cloning feature to the service so that users can answer each answer in the voice of the presidential candidate.

I’d long wanted to experiment with Lovable AI, an AI-powered web-app builder that turns natural-language prompts into deployable web applications, not requiring traditional web-dev skills. We chose to build our service with with Lovable, and remarkably, we had a fully functional version live within just a few hours. The app uses OpenAI’s ChatGPT API to generate responses, and uses ElevenLabs to clone the voices of the candidates.

Let me share the demo clip below (it is Korean service). I compared with the candidates’ real speeches.

Lovable turned out to be a surprisingly powerful tool for spinning up proof-of-concept demos at minimal cost. It really feels like we are entring a new era, one where trying out new ideas and prototyping services is faster and more accessible than ever before, even for small teams.

Source code is available on Github.