Inspiration

Nigerian university students share countless study materials in chaotic WhatsApp groups—blurry handouts, badly lit photos of physical past questions, and illegible handwritten notes. We wanted to build a platform that doesn’t just read these messy images, but acts as a dedicated, interactive tutor to break them down.

What it does

OpenSolve NG is an open-source multimodal learning engine. Students can upload photos of past questions, PDF syllabuses, or code snippets directly into the workspace. Using the Gemma 4 Multimodal API, the platform extracts the text/math, reasons through the logic, and generates a formatted, step-by-step explanation. Users can then chat directly with the AI to ask follow-up questions in real-time.

How we built it

We built a responsive, mobile-first frontend using Next.js (App Router) and Tailwind CSS. The core intelligence is powered entirely by the Google Gen AI SDK, specifically leveraging the gemma-4-26b-a4b-it model for its incredible open-weights multimodal capabilities. We engineered a custom API route that safely handles multipart form data (images and PDFs) alongside dynamic chat history, ensuring seamless continuous conversation without dropping context.

Challenges we ran into

Handling high-resolution image uploads alongside text history on mobile devices initially caused Vercel payload limits and UI clipping. We overcame this by implementing strict responsive fluid wrappers (min-w-0, overflow-x-hidden) and sanitizing the multipart form data buffer strictly before passing it to the Gemma API.

Accomplishments that we’re proud of

We successfully integrated a pure Gemma 4 multimodal pipeline without relying on closed-source models. The platform handles complex mathematical reasoning, messy handwriting OCR, and continuous chat memory in a single, fluid UI that rivals commercial platforms.

What we learned

We learned how to orchestrate multimodal prompts, manage dynamic API buffers in Next.js, and force CSS Flexbox to respect mobile bounds when rendering dynamic Markdown tables and code blocks.

What’s next for OpenSolve NG

We plan to expand the open-source repository so developers from different universities can inject custom prompts.json files tailored to their specific departmental grading rubrics.