Project ideas from Hacker News discussions.

Show HN: Prompt-to-Excalidraw demo with Gemma 4 E2B in the browser (3.1GB)

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Three dominant themes inthis discussion

Theme	Key take‑aways (with direct quotes)
1. Browser and WebGPU limitations, especially in Firefox	“*Firefox has WebGPU already, but the subgroups* extension isn’t in yet. Every matmul / softmax kernel here leans on `subgroupShuffleXor` for reductions, that’s the blocker.*” – teamchong*
2. Need for smarter model caching / sharing across sites (CDN, P2P, or browser‑level cache)	“...would be great if there was a way that I don’t have to redownload them across demos so that I just have a cache. or an in‑browser model manager.” – hhthrowaway1230 “CDN wouldn’t help much. These days browsers partition caches by origin, so if two different tools fetch the same model, the browser would download it twice.” – wereHamster “I built a temporary CDN … `https://stateofutopia.com/experiments/ephemeralcdn/`” – logicallee
3. Browser‑specific performance constraints (batch‑size 1, memory bandwidth, security)	“Small models in the browser are a different optimization problem than small models on a server. On server you chase throughput so you batch. In browser you’re stuck at batch size 1, which means kernel launch overhead and memory bandwidth dominate, not FLOPs.” – osamaJaber “The Gemma models really are amazing. I was on a flight … used E2B to run the model locally on my Pixel 10 Pro.” – walthamstow

These three topics capture the most frequent concerns and suggestions voiced by participants in the Hacker News thread.

🚀 Project Ideas

Generating project ideas…

WebModelCache#Summary

Browser extension + lightweight API that lets any webpage store and retrieve large ML model binaries from a shared IndexedDB cache, bypassing repeated downloads.
Solves the “download the same 2 GB model every time I visit a new demo” frustration while preserving privacy.

Details

Key	Value
Target Audience	Web developers and hobbyists running browser‑based LLM demos who repeatedly download models
Core Feature	Cross‑origin, hash‑based shared cache that automatically re‑uses model files already present in any site’s IndexedDB
Tech Stack	Service Workers, IndexedDB, Web Crypto (SHA‑256 hash), BroadcastChannel, Manifest V3 extension
Difficulty	Medium
Monetization	Hobby

Notes

HN users repeatedly asked for a “cache so I don’t have to redownload models across sites” – this provides it.
Practical utility: cuts bandwidth usage, speeds up demo loading, works offline, and can be bundled as a plug‑and‑play extension.

SubgroupBridge

Summary

A cross‑browser polyfill that emulates WebGPU subgroup operations (e.g., subgroupShuffleXor) for browsers lacking native support.
Enables existing LLM inference kernels that rely on these extensions to run on Firefox and older GPUs.

Details

Key	Value
Target Audience	WebGPU developers, LLM inference hobbyists, Firefox users
Core Feature	Runtime translation of subgroup API calls into WebAssembly kernels that simulate reductions
Tech Stack	TypeScript, WebAssembly, WebGPU API shim, Service Worker for fallback kernels
Difficulty	High
Monetization	Hobby

Notes

Team members explicitly said “subgroupShuffleXor is the blocker for Firefox WebGPU” – this removes the blocker.
Potential discussion hook: how to safely expose emulated subgroups without compromising security.

EphemeralCDN

Summary

A temporary ultra‑fast CDN for model assets combined with built‑in WebTorrent peer‑to‑peer fallback, so downloads happen once and are shared across sites and users.
Eliminates the need to re‑download large models for every demo or experiment.

Details

Key	Value
Target Audience	LLM demo creators, educators, and hobbyists who host interactive model demos
Core Feature	One‑click CDN URL that serves files at edge speed; if a peer already has the file, it streams via WebTorrent to reduce origin load
Tech Stack	Cloudflare Workers, Next.js front‑end, WebRTC/DataChannel (WebTorrent), Service Worker caching
Difficulty	Medium
Monetization	Revenue-ready: Freemium (free tier up to 10 GB/month, $5/mo for higher bandwidth)

Notes

Directly answers the “CDN or uberfast downloader” request and the desire for “upload at the same time” peer model.
High practical utility: reduces data costs for users on limited connections and enables offline reuse of models across domains.