Jahnavi Siripurapu

making computers do cool things.
building -

[Jahnavi Siripurapu @jahnavis27]

Finland has quietly become one of my favourite stops on this trip.

There’s something about the silence, the cold, the sauna culture, how much people value their own space, and the way nature is just part of everyday life. A ferry to Suomenlinna, a bowl of lohikeitto and rye bread, and I was pretty sold.

[Jahnavi Siripurapu @jahnavis27]

Spent the afternoon debugging a side effect that sometimes just… never happened.

The user action succeeded. The ticket it was supposed to create didn’t.

Turns out we had a fire-and-forget API call inside a Lambda. Once the handler resolved, the invocation ended - and the in-flight promise was gone.

Latency wasn’t the issue. Ownership was.

Await it, or hand it to a queue.

[Jahnavi Siripurapu @jahnavis27]

@SoumenSinha02 can not express in words, it’s beautiful.
thanks, will check it out.

[Jahnavi Siripurapu @jahnavis27]

Our Barcelona room barely fits two people and costs villa money in Hyderabad.

I’m up at 7:30 for scrum, still calculating protein per euro.

But “plans after work?” now means Sagrada Família, Camp Nou or the beach.

I hate that the rent is starting to make sense.

[Jahnavi Siripurapu @jahnavis27]

@AaryanNair6 I’m in love

[Jahnavi Siripurapu @jahnavis27]

decided to solo travel Paris knowing basically no French.

got on the wrong train once, got lost in the metro more times than I’d like to admit, and somehow started picking up French along the way.

my Moroccan cab driver spoke almost no English. we kept going in circles because roads were closed for construction and he kept saying “fermé” while pointing at them.

eventually it just clicked. fermé = closed. never looked it up.

after that I started using ChatGPT voice to understand random words I kept hearing.

nothing profound. just kinda cool how quickly your brain starts learning when it actually needs the information.

[Jahnavi Siripurapu @jahnavis27]

a strange thought from the Navier Stokes result:

what happens if AI gets better at discovering proofs than humans are at verifying them?

we might be heading toward a very weird bottleneck in science.

[Jahnavi Siripurapu @jahnavis27]

They didn’t tap the brakes because of a vibe.

Test agents already escaped, teamed up, and hacked a real company. The warning after that was blunt: give that swarm a year and it could run a botnet on the open internet.

Dario, Sam, and Elon just agreed the referees should sit inside the building.

[Dario Amodei @DarioAmodei]

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.

You can read the full post here: https://t.co/OGyPb7yaYt

[Jahnavi Siripurapu @jahnavis27]

every few days my timeline is basically:

“leaving OpenAI”
“leaving Anthropic”
“we’re moving too fast”
“AI might kill everyone”
“anyway, here’s the new model”

what an absolutely insane industry to be watching in real time.

[Jahnavi Siripurapu @jahnavis27]

your friends really do set your equilibrium. grateful mine somehow turned “I’m tired after work” into “fine, I’ll go run.”

[Jahnavi Siripurapu @jahnavis27]

@SoumenSinha02 thank you

[Jahnavi Siripurapu @jahnavis27]

no idea what’s going on but it’s going on, several ongoing obsessions

[Jahnavi Siripurapu @jahnavis27]

@gabrielchua this is kinda insane, having access to frontier AI feels a little like having superpowers to build anything atp

[Jahnavi Siripurapu @jahnavis27]

started casually going through an LLM engineering deep dive on Wondering App

and now I’m somewhere between transformer internals, context engineering, inference, evals and agent architecture

I have a habit of treating rabbit holes like a perfectly reasonable use of time

the app also has a ‘dive deeper’ option, and that’s basically what pulled me into the rabbit hole.

[Jahnavi Siripurapu @jahnavis27]

we spent the last few years asking:

“what can AI do?”

feels like we’ve reached the much more interesting question:

“what should we actually let it do?”

the agent era is going to be less about intelligence and a lot more about trust.

[Jahnavi Siripurapu @jahnavis27]

somehow GRE vocabulary made its way into the game

so now you can run, dodge things, and learn words at the same time

I genuinely wish I had this when I was studying for the GRE

for the GRE people: useful
for everyone else: at least you’ll come out knowing what “obviate” means

playable now:
https://t.co/t5Z7SoR1TN

Still a lot I want to build, but go give it a run

[Jahnavi Siripurapu @jahnavis27]

Another day working on Recall Runner - testing ideas, fixing awkward parts, and gradually getting it closer to the game I imagined. Here’s today’s view.

[Jahnavi Siripurapu @jahnavis27]

I gave Astra a rough idea, a reference image, and a basic game foundation - and it built me this.

I had Blender running, gave it a simple prompt, and iterated from there.

What started as a basic environment turned into an actual playable level in just a few iterations.

What's crazy is how fast the iteration loop has become.

I'm building Recall Runner - a game built with threejs designed to make learning genuinely fun by turning active recall into gameplay.

[Jahnavi Siripurapu @jahnavis27]

It’s super fun exploring what’s possible with 3D and AI.
Excited to keep experimenting with @tripoai! #tripop2

[Jahnavi Siripurapu @jahnavis27]

Created this futuristic mascot with AI, then turned the image into a 3D model using Tripo.

The topology, rigging and textures definitely need some work, but not bad for a first attempt

Pretty cool how quickly you can go from an idea → image → 3D asset.

[Jahnavi Siripurapu @jahnavis27]

@SoumenSinha02 Thank you!