The Chatbot That Said 'No' — And How I Found the Right Words for Local Dialect
When building Local Dialect, the chatbot refused to respond to certain questions, forcing a deeper exploration into language learning design. This session with Claude revealed the need for a more culturally relevant approach, ultimately shaping the platform's unique learning experience.
I was working on the chatbot feature for Local Dialect, a language learning platform focused on Kenyan and African languages. The goal was to create a conversational AI that could help users practice speaking and listening in Swahili and regional dialects. I asked Claude to help draft the chatbot’s initial responses, but the AI kept rejecting the prompts, returning messages like, 'I don’t know how to respond to that.' It was frustrating — I had a clear vision for what the chatbot should do, and it wasn’t aligning with what the AI was suggesting.
Claude pointed out that the chatbot was struggling with the nuances of language learning. The prompts I was using were too generic — they didn’t account for the specific cultural and linguistic context of Swahili and regional Kenyan dialects. The AI wasn’t just rejecting the prompts; it was highlighting a deeper issue: the chatbot needed to be more than a generic question-answering tool. It needed to be a guide, a companion in the learning process.
I revisited the design with Claude and Codex, focusing on how chatbots are used in language learning. We brainstormed scenarios where the chatbot could act as a patient tutor, offering feedback, corrections, and encouragement. We even tested this approach by writing sample dialogues where the chatbot would ask open-ended questions and provide tailored responses based on the user’s level and progress.
What finally worked was a shift in how I framed the chatbot’s role. Instead of making it an automated response system, I designed it as a learning partner. The chatbot would start with simple greetings, then move into sentence construction, and eventually offer feedback on pronunciation and grammar. The AI helped me structure this flow, ensuring that each interaction felt natural and supportive.
The result was a chatbot that users could actually engage with — one that didn’t just repeat information but encouraged them to practice and improve. It wasn’t just a technical win; it was a cultural one. The chatbot now reflects the way Swahili and regional languages are spoken in everyday life, making the learning experience more authentic and relatable.
Building AI-powered tools for everyday life in Kenya isn’t always about getting the right answer — sometimes it’s about asking the right questions. And sometimes, the AI is the one who helps you find them.
If you're interested in seeing how AI can be used to build practical tools that make a real difference, subscribe to the newsletter and get insights like this delivered to your inbox every Monday and Friday. [Subscribe here](https://wambugumartin.com/subscribe).
