Apoorva Joshi

Apoorva Joshi

@apoorvaj.io

Computer graphics, systems programming, general methods. Working on graphics at Unity. Formerly Playdead, NVIDIA, Microsoft. apoorvaj.io 📍Copenhagen

880 followers214 following76 posts

Apoorva Joshi reposted

This is effectively Gaussian Splatting just without the Gaussian part.

I'm working on vector editing in @papaya.io, and it's not as snappy as I'd like to be. So optimization time! I built a profile capture mechanism: An in-browser shortcut toggles profiling, then pings the local server which converts it into the Tracy format, writes to disk, which I view in the GUI.

Impressions on DeepSeek V4 Pro 0813: Really great model. Asks clarifying questions, pushes back as appropriate, strikes a good balance between eagerness and parsimony. And it is supposedly ~39x cheaper than Opus 5.

I don't know what to think of benchmarks any longer. DeepSeek V4 Flash 0731 also scored fantastically on benchmarks and shot through Pareto frontier, but got stuck in thinking chains in my testing more than once.

This is a super fun feature. You can just paste in most ShaderToy effects. What's a better way to add bling to your graphics programmer CV than literally add a shader in there, and then just export to PDF? 😅

papaya.io@papaya.io

papaya.io now has shader layers, so you can create all kinds of custom effects. Supports a whole bunch of ShaderToy shaders out of the box, plus you can declare uniforms to get UI controls like sliders or color selectors. Also supports multi-pass effects like Gaussian blurs.

Apoorva Joshi reposted

openai announces that in a worrying demonstration, during a recent benchmark exercise requiring models to solve a math conjecture, their latest model broke out of a sandbox and gained illicit backdoor access directly to the platonic realm of forms

Apoorva Joshi reposted

At #SIGGRAPH2026, I presented "points as tori" (PAT) for fast pointwise evaluation of signed distance for point clouds. Some code (Python w/ C++ extensions) and demo up: github.com/nzfeng/point... Contributions from people/LLMs welcome. Explainer thread below for the project: 🧵

Apoorva Joshi reposted

Hey everyone. A reminder that the deadline for graphic article proposal for GPU Zen 5 is approaching quickly. Sharing knowledge is the best way to learn!

🎮Graphics Programmers! Ready to shape the future of advanced rendering? Submit your proposal for GPU Zen 5, the next volume in the acclaimed GPU ZEN series, today! 📅 Deadline: July 5, 2026 📩Submit to: gpuzen@theforge.dev 🔗 gpuzen.blogspot.com #GraphicsProgramming #Rendering #GameDev #Authors #GPU

Apoorva Joshi reposted

papaya.io ships new features all the time 🚀 Help us prioritize what comes next. You can now drop feedback right from the toolbar on the left. Takes 10 seconds. We read every word.

Claude Fable has a certain clarity of thought in English. It says stuff that sounds like a smart person, not a compressed AI shoggoth. Interestingly, it still isn’t very good at API taste.

Still having fun building papaya.io. Just implemented a pretty nifty compressed export feature, with swipe to compare. Supports JPEG, WebP and PNG compression for now.

I'm trying out zed.dev since they announced v1.0. It feels like a real labor of love, and had no rough edges in my first day of using it. Comes with batteries included, and things Just Work. Feels more suited to my agent-heavy workflow than VS Code.

I had an index variable that kept track of the latest processed element in a list. gpt-5.3-codex today named it `watermark`, and I cannot get over the beauty of it.

Apoorva Joshi reposted

Qaws (قوس) Arabic for *arc*, *bow*, *curve*. A dependency-free C11 library for creating, evaluating, sampling, traversing, and inspecting parametric curves and surfaces in 2D and 3D. A dependency-free C11 library for parametric curves and surfaces in 2D/3D. 1/13

This was also my experience with Steve Yegge's Gas Town (steve-yegge.medium.com/welcome-to-g...). Total jank, but there's a kernel of a really powerful idea in there.

steve-yegge.medium.comWelcome to Gas TownHappy New Year, and Welcome to Gas Town!
Apoorva Joshi@apoorvaj.io

Tried OpenClaw to eval the hype. It's janky af; couldn't get it to work in an hour of troubleshooting in Docker. Pure vibes coding is still a trash fire, but someone's going to build a very smooth version of this, and it's going to be 🔥.

Tried OpenClaw to eval the hype. It's janky af; couldn't get it to work in an hour of troubleshooting in Docker. Pure vibes coding is still a trash fire, but someone's going to build a very smooth version of this, and it's going to be 🔥.

I feel that Claude Code + Sonnet writes way more tasteful Zig code than it does JavaScript. JS is littered with dumb defensive programming, with null checks or conditions where here should be asserts. My theory is that this is because of the delta in the quality of Zig vs JS human training data.

This was an easy one to implement. Then I started on white balance, fell into a rabbit hole of black body radiation and color spaces, and then noped out for a bit.

This post is unavailable.

I'm really proud of this. 1. Unlike many other web-based image editors, papaya.io rasterizes text in WebAssembly. This means that text rendering looks the same across browsers or OSes. 2. This means that I had to re-implement a text box, which was not fun. ...

papaya.io@papaya.io

After three months of work, text editing is now live! papaya.io

3. This is where @mikkomononen.bsky.social 's excellent Skribidi library came to the rescue. This might be the first WASM implementation of Skribidi in the wild. I was able to make some upstream contributions. 4. Papaya supports Arabic, Chinese-Japanese-Korean, Hebrew, Latin and 20 more scripts.

5. Papaya uses Harfbuzz and FreeType for high quality rendering. 6. Text rendering is done in a worker WASM thread. So it is async and faster than JS. Text rendering is hard, and literally half of all commits to papaya.io so far. So please let me know when you find bugs! :)

*Tin foil hat* Has Google added a special case for Google Docs in Chrome to show the MacOS accent menu in the right spot? It renders out of the window in Firefox, which is what you'd expect to happen when using a hidden textbox to get input. I cannot find a web API do this. Looks like an inside job.

If you've ever worked with font rendering, then you'll know that it's the most impressive to get the fonts looking good at low sizes, because then hinting comes into play, which is where FreeType shines over other simpler libraries.