‎## Inspiration ‎Driven by the healthcare disparities in rural Nigeria and the Global South, we wanted to build an AI that doesn't rely on the internet to save lives. Combining molecular biology (CRISPR) with local AI inference was the ultimate goal. ‎ ‎## What it does ‎Vision-Link AI is an offline diagnostic assistant. It helps health workers analyze molecular data and CRISPR-based diagnostic results locally on AMD-powered hardware, providing instant medical insights without needing a cloud connection. ‎ ‎## How we built it ‎We utilized the Qwen2.5-0.5B model, optimized for local inference. The backend is built with Python, while the frontend uses a modern Next.js interface. We integrated specialized bioinformatics logic to handle CRISPR research data. ‎ ‎## Challenges we ran into ‎Optimizing a large model to run offline on consumer-grade hardware was tough. We also faced cross-platform technical issues, like Git directory path errors (trailing spaces in folder names) that broke our Windows vs. Linux collaboration! ‎ ‎## Accomplishments ‎Successfully deploying a functional AI Agent that can provide diagnostic support 100% offline. Coordinating a global team across 4 countries to build a life-saving tool in record time.

‎## What we learned ‎We gained deep insights into optimizing Small Language Models (SLMs) like Qwen2.5 for offline edge computing. We also learned the critical importance of cross-platform version control (Git) when collaborating across different operating systems (Windows vs. Linux) and the immense potential of AI in democratizing CRISPR diagnostics. ‎ ‎## What's next for Vision-Link AI ‎Our next step is to expand the hub into a Multimodal System (Track 3) that can interpret medical scans and X-rays offline. We also plan to fine-tune our models on larger, localized African medical datasets (Track 2) and partner with rural health NGOs to pilot Vision-Link AI Hub in remote clinics across Northern Nigeria. ‎

Built With

  • amd-rocm
  • biotechnology
  • crispr
  • git
  • hugging-face
  • nextjs
  • python
  • qwen2.5-llm
  • tailwind-css
Share this project:

Updates