Replacing the generated “person behind the name” text with a real dataset is the rebuild I respect most here. Like you said, generated copy reads smoothly and means nothing. I feel the same pull building an AI resume tool: it’s tempting to let the model fill every gap, but people trust what they can check. And the voice clip being the most-mentioned feature is a great reminder that the smallest thing is often what people remember. Looking forward to the technical write-up on the multi-language pipeline.