DugAssistant

Inspiration

The story of DugAssistant begins at home, with our father, Bacibone Metre Dug, a dedicated math teacher with over 30 years of experience. We grew up witnessing both his passion for teaching and his struggles: sleepless nights preparing lessons, frustration over outdated materials, and a constant effort to do his best with little support. This daily grind took a toll on his energy and well-being. Seeing this, we felt compelled to act not just for him, but for countless teachers facing similar challenges.

What it does

DugAssistant is an AI-powered assistant designed to make lesson planning easier and more efficient. Tasks that once took hours like writing lesson plans, finding exercises, and creating quizzes now take less than 10 minutes. Teachers can use images or audio inputs throughout the preparation process to generate creative and immersive lessons.

How we built it

DugAssistant is a React Native application built with Expo. It uses the OpenAI API to generate content and understand images. It also leverages ElevenLabs to convert speech to text for audio inputs. In addition, it uses OpenAI's RAG (Retrieval-Augmented Generation) functionality to process PDF documents, providing context and references to the generated lessons. All these features are connected to Supabase functions, where additional logic is handled, along with a custom-deployed service that exports lessons to Word documents. Most of this was generated using Bolt and carefully crafted prompts.

Challenges we ran into

One of the biggest challenges we faced while building DugAssistant was simplifying the user experience while simultaneously reducing AI token consumption. DugAssistant supports multiple teaching methodologies, which required us to work closely with teachers to design an optimal lesson creation process. Due to the large size of the app, Bolt would sometimes replace correct sections with incorrect ones. This issue was mitigated through the use of GitHub version control.

Accomplishments that we're proud of

The application can process text, audio, and images while offering teachers a choice of methodologies. Compared to our first version, the current app is more dynamic and involves the user more directly in the lesson generation process. With this, DugAssistant can scale more effectively and help teachers become more productive in more countries.

What we learned

Throughout this experience, we significantly improved our prompt-writing skills. Our final prompts were much more efficient and results-oriented than the earlier ones. We also learned to narrow the context and be more explicit in our instructions, leading to better results. Additionally, we were introduced to Supabase (functions, database, and authentication), which we found to be powerful and interesting.

What's next for Dug Assistant

With our first version last year, we served around 700 teachers across two pilot regions in the DRC. Based on the feedback we collected, we designed and built this new version using Bolt. Our plan is to deploy it across the entire DRC (approximately 350,000 teachers) and expand to neighboring countries before eventually scaling across the continent and beyond.