Many small clothing sellers in Nigeria use WhatsApp Status to display their products. The most tedious part of this process is writing a unique caption for each item. This project, CaptionBot, aims to automate this task. Sellers can describe their items through typing or voice, and CaptionBot generates a ready-to-post WhatsApp caption. The goal is a small, focused solution for a specific problem, not a general AI assistant.The CaptionBot application is lightweight and runs on Python, utilizing Gemma 3 1B locally for caption generation. Chrome's voice typing allows for spoken input, with a preprocessing step to correct transcription errors. Gemma handles the language, while custom code enforces rules and validates generated content. The choice of a small, local model minimizes data costs for sellers.A significant challenge encountered was the model inventing details, such as a price, that were not provided in the input. This occurred due to misleading examples in the prompt. Transcription errors also presented issues, with some words being misinterpreted. These problems were addressed by refining the prompt, removing misleading examples, and adding code-level validation for critical information like prices.The project emphasizes open innovation, allowing for direct control over the model's behavior and the application's logic. This approach avoids per-use API costs associated with hosted models. The next crucial step involves user testing with actual sellers to gather feedback on preferences, language handling, and other aspects of their workflow. This feedback will shape CaptionBot into a tool truly designed by its users. The prize category recognizes its innovative use of Gemma for local, practical problem-solving.