i made a prompt that generates premium open-graph hero images like a real brand designer

tested it on major brands, all came out clean!

one prompt. any brand. 16:9, zero noise, SaaS-grade visuals.

drop it into Gemini → instant marketing asset 👇

gist.github.com/devarshishim...

i made a prompt that generates premium open-graph hero images like a real brand designer

tested it on major brands, all came out clean!

one prompt. any brand. 16:9, zero noise, SaaS-grade visuals.

drop it into Gemini → instant marketing asset 👇

gist.github.com/devarshishim...

pricing is fully hardcoded in modelCost.ts and there's a fast-mode called "penguin mode" that costs 6x Opus 4.6.

that's a lot of tokens.

pricing is fully hardcoded in modelCost.ts and there's a fast-mode called "penguin mode" that costs 6x Opus 4.6.

that's a lot of tokens.

Claude Code's source just leaked on npm via an accidental .map file push.

~512K lines.
~1,900 files.
Full app structure.

No hack involved. Just a bad deploy.
github.com/instructkr/c...

Claude Code's source just leaked on npm via an accidental .map file push.

~512K lines.
~1,900 files.
Full app structure.

No hack involved. Just a bad deploy.
github.com/instructkr/c...

𝐀𝐈𝐓𝐚𝐛 𝐢𝐬 𝐧𝐨𝐰 𝐅𝐥𝐨𝐰𝐓𝐚𝐛!⚡

Same fast new tab. Same zero bloat. New name that actually fits.

FlowTab — NewTab, Search & Focus. Open GitHub, Figma, and Slack instantly. FlowTab brings universal app search and intelligent shortcuts to your browser’s core.

Your browser deserves better than the default.

Install now!
flowtab.devarshi.dev

𝐀𝐈𝐓𝐚𝐛 𝐢𝐬 𝐧𝐨𝐰 𝐅𝐥𝐨𝐰𝐓𝐚𝐛!⚡

Same fast new tab. Same zero bloat. New name that actually fits.

Your schema will cost you more than your server ever will.

I learned this building an internal tool. Search suggestion cache on Postgres. Clean schema, proper indexes. At 25k rows, it felt solid.

Then I modeled 1M queries x 250 regions.
250 million rows.

Devarshi Shimpi - Founder, Software Engineer. I'm Devarshi, founder of PrismLink, building production-ready apps and empowering developers through open-source and DevOps expertise.

I write about stuff like this in my newsletter. No spam. Just useful tips and tricks to help you grow as a developer.

500+ developers already in. Join here: go.devarshi.dev/newsletter

B-tree index: 12GB. Heap: ~330GB. $85/month for one table.

250M rows to 1M rows. ~$85/month to ~$0.50/month.
The data didn't change. The structure did.

Before your next CREATE TABLE, ask what data is actually unique per row. Everything else is a candidate for collapsing.

"javascript tutorial" stored 250 separate times. Suggestions 95% identical. Full payload duplicated every time.

The fix was a JSONB collapse. One row per query, regional keys inside the JSON, fallback to "default" for identical regions.

Your schema will cost you more than your server ever will.

I learned this building an internal tool. Search suggestion cache on Postgres. Clean schema, proper indexes. At 25k rows, it felt solid.

Then I modeled 1M queries x 250 regions.
250 million rows.

React Server Components are powerful, but "React2Shell" proved that power comes with a massive attack surface.

WAF blindness. DoS loops. Source leaks.

I broke down exactly how the exploit works and why your firewalls missed it.

Link below:👇

React (RSC) Exploits Are Real and It's Hurting Engineering Teams Deep dive into React RCE (CVE-2025-55182) & Next.js risks. Covers RSC Flight exploits, patch fatigue, and securing server actions from DoS & leaks.

Link: devarshi.dev/blog/react-r...

React Server Components are powerful, but "React2Shell" proved that power comes with a massive attack surface.

WAF blindness. DoS loops. Source leaks.

I broke down exactly how the exploit works and why your firewalls missed it.

Link below:👇

AITab onboarding is live!🎉

Try now:👉https://aitab.devarshi.dev

DigitalOcean: AI-Powered Unified Agentic Cloud Infrastructure Build on DigitalOcean's unified agentic AI cloud infrastructure. AI-powered development, instant deployment, easy management. Simple and affordable.

Funding my experiments one referral at a time 😅. Here is $200 credit if you're new to DO: m.do.co/c/645d44d1a7a1

Windows -> VirtualBox -> Ubuntu desktop -> DigitalOcean Ubuntu server -> Windows.

Cuz DigitalOcean officially doesn't allow running Windows Droplets

Windows -> VirtualBox -> Ubuntu desktop -> DigitalOcean Ubuntu server -> Windows.

Cuz DigitalOcean officially doesn't allow running Windows Droplets

I thought my code was perfect, until a user tried it.

I recently pushed an update to AITab and felt pretty confident. But shortly after, my friend pointed out something I had completely missed something.

He was spot on. The UI was flickering.

The result?

✅ Zero flickering: The UI stays stable until the user is ready.
✅ 90% fewer API calls: No more wasted bandwidth on partial queries.
✅ A happier userbase: (And a much smoother experience).

You can try the new, smoother search yourself here: aitab.devarshi.dev

Every time you type a letter, the interface jumped. Type "H" → Request sent. Type "He" → Request sent. Type "Hel" → Request sent.

I fixed it by implementing a 300ms debounce. Now, the app waits for the user to pause before fetching data.

I thought my code was perfect, until a user tried it.

I recently pushed an update to AITab and felt pretty confident. But shortly after, my friend pointed out something I had completely missed something.

He was spot on. The UI was flickering.

Recreated the iconic Vercel Ship 3D card for my portfolio!⚡️

Always loved the interactivity of the original, so I had to build my own version. The floaty physics are so satisfying to play with.

Check it out:👉 devarshi.dev

Recreated the iconic Vercel Ship 3D card for my portfolio!⚡️

Always loved the interactivity of the original, so I had to build my own version. The floaty physics are so satisfying to play with.

Check it out:👉 devarshi.dev

Left or right?

Btw incase you are curious about the product!

bsky.app/profile/deva...

Left or right?