# Cache Me If You Can

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Shubhranshu Sudeepta Panda](https://cerebralvalley.ai/u/shubhranshu_)
- **GitHub:** https://github.com/shubhranshu-p/napkincloud
- **Demo video:** https://youtu.be/WKyKA6PkrjU
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5237

NapkinCloud — Sketch. Test. Improve.

NapkinCloud is a Meta-powered architecture workbench for founders and backend teams. It turns a system diagram into an editable, reproducible traffic experiment so builders can identify bottlenecks and compare a tested repair against the original.

Upload a hand-drawn or digital architecture diagram, review the extracted components and connections, and configure traffic, service capacities, queues, deadlines and performance targets. Run a baseline, then let Meta propose bounded changes such as eligible-read caching, API replicas or independent tenant database partitions. Each candidate runs against the same workload and random seed, making before/after comparisons meaningful.

The workbench shows animated request flow, latency and failure metrics, bottleneck findings, stress cases, an explanation and a generated architecture illustration. Users can export evidence and reopen browser-local saved runs. Three sample SaaS cases and a deterministic simulator work without an API key.

The key design principle is that the simulator determines the numbers: model prose and generated artwork cannot set performance results. Failed repairs remain visible, added resources and assumptions are explicit, and a separately labeled bounded verification search can test partition counts when model candidates are insufficient.

Built with JavaScript and a Node.js local backend, with a Cloudflare Worker deployment target. API credentials stay server-side. Results are simulations, not production benchmarks; the tool helps test architecture hypotheses before committing to infrastructure changes.

## More from Meta Global AI Developer Hackathon

- [KREO_ONE](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5234)
- [LavaFlow AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5235)
- [AMADEUS](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5236)
- [Sourcebound](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5238)
- [AI Finds, Lean Checks](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5239)
- [Helpious AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5240)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5237. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt