Challenge
Build a PoC That Benefits Society
Build a proof-of-concept solution that you think solves a real-world problem. You have creative freedom to tackle any challenge you want.
Judging Criteria
Projects are evaluated across two equally weighted dimensions:
- Solution Quality (50%): How effectively your PoC solves the chosen problem, how functional and demonstrable it is, and how meaningfully does it address XAI or computer vision problems.
- Real-World Applicability (50%): How relevant the problem is today, whether the solution could be deployed in practical scenarios, and whether there is a clear path from PoC to production.
Inspiration and Ideas
Consider domains where understanding LLMs matters:
- Healthcare: Explaining medical advice or diagnosis assistance.
- Legal and compliance: Interpreting contract analysis or regulatory guidance.
- Education: Understanding how AI tutors arrive at explanations.
- Content moderation: Explaining why content was flagged.
- Financial services: Making credit decisions or fraud detection interpretable.
- Customer support: Understanding chatbot response generation.
Consider domains where vision-language modeling matters:
- Accessibility: Better scene and object descriptions for assistive tools.
- Healthcare imaging workflows: Safer clinician support.
- Industry and logistics: Interpreting visual inspection and anomaly reports.
Submission Guidelines
- Open source: Publish your code on GitHub.
- Working PoC: The repository should include an implemented proof-of-concept or design.
- submission.md: Include a
submission.mdfile in your repository with a concise write-up covering the problem statement, solution approach, along with the application and expected real-world impact.
Tips for Success
- Start simple: Pick a focused problem you can realistically solve or design in a weekend.
- Think users, not features: Be explicit about who benefits and how.
- Show, do not tell: A working demo is better than a perfect pitch.
- Answer the “so what?”: Explain why your solution matters for your use case.
- Be realistic: Prioritize current, practical problems over hypothetical ones.
Rules
- Build or design something that works well enough to demonstrate.
- You can use any programming language, framework, or additional tooling.
- Individual and team submissions are accepted.
Resources
Support is available on Discord.
The strongest submissions will not just run; they will show why interpretable AI matters for real people and real decisions.