Priya Roul

I build backend systems at scale and AI agents.
Engineer @gocashfree · GSoC ’25 @SwiftLang
Active in AI, developer tools & distributed systems, art and dance!

[Priya Roul @priyaaroul]

prototype worked with waveshare esp32..
final outcome very soon🤓

[DHH @dhh]

Burnout happens when you work too hard with not enough meaning.

"He who has a why to live for can bear almost any how" — Nietzsche

[Priya Roul @priyaaroul]

@palak_agrawalll konse city me ho?

[Priya Roul @priyaaroul]

hii bangalore girlies, I will be hosting a dance workshop soon in blr

dance style: semiclassical

DM me! if you are into it, it’s going to be super fun and lots of cardio

[Priya Roul @priyaaroul]

@bekhayaalii nooo, this will be next sunday

[Amitosh @recrsn]

1.5 years ago I got a 3d printer to “print stuff”, @priyaaroul makes actual art in it.

[Amitosh @recrsn]

Our house is adorned with many 3d mandalas and little items made by @priyaaroul

[Priya Roul @priyaaroul]

@recrsn 🥹🥹

[Priya Roul @priyaaroul]

made this cutest bookshelf, will add more Harry Potter themed miniatures in the next iteration

[Priya Roul @priyaaroul]

@takeiteasyUrvy 3d printed the shelf, books from Amazon. The books are incredibly cute.

[prerak @prerak_011]

One day I'm gonna put my foot down and decide that my youth is too precious to waste away in Bangalore traffic

[Priya Roul @priyaaroul]

@heykavsthere more power to you

[Priya Roul @priyaaroul]

a good 10k run💪

[Priya Roul @priyaaroul]

@deevee47 This is a little on the sour and sweet side, if you like this taste note, you can try!

[Priya Roul @priyaaroul]

This is the best coffee packaging I've received, and I am so excited to try it!

It has a caramel note and is a light roast with a little sweetness.

[Priya Roul @priyaaroul]

I am always a big fan of pour-overs and go crazy for monsoon-malabar coffee beans.

Recently, we have been trying different coffee beans every month curated by aaramse, and it's been interesting to judge the tasting notes and share feedback.

We received our beans for sept, let's see how this goes.

[Priya Roul @priyaaroul]

Don't hand over all your engineering decisions to an LLM.

Use it as a collaborator instead.

Before you implement, have it challenge your plan. I use skills like /grill-me to find gaps, refine the approach, and only then start coding.

A few things that have saved me a lot of time:

- add your coding preferences to your local LLM rules: no unnecessary comments, follow the correct Liquibase style, stay consistent with existing classes and patterns.
- every time the LLM does something you don’t want, don’t just correct it once. Add that lesson to a rule or script it can refer to next time.
- build a simple test framework around the features you’re working on, so the LLM can implement → run tests → fix issues in one loop.
- review the plan before reviewing hundreds of lines of generated code. Fixing the direction early is much cheaper than fixing a messy PR later.

The goal isn’t to make the LLM write more code.

It gives it enough context, rules, and feedback loops that it needs less supervision each time.

This has saved me a surprising amount of time, tokens, and rework.

[Priya Roul @priyaaroul]

Don't hand over all your engineering decisions to an LLM.

Use it as a collaborator instead.

Before you implement, have it challenge your plan. I use skills like /grill-me to find gaps, refine the approach, and only then start coding.

A few things that have saved me a lot of time:

- add your coding preferences to your local LLM rules: no unnecessary comments, follow the correct Liquibase style, stay consistent with existing classes and patterns.
- every time the LLM does something you don’t want, don’t just correct it once. Add that lesson to a rule or script it can refer to next time.
- build a simple test framework around the features you’re working on, so the LLM can implement → run tests → fix issues in one loop.
- review the plan before reviewing hundreds of lines of generated code. Fixing the direction early is much cheaper than fixing a messy PR later.

The goal isn’t to make the LLM write more code.

It gives it enough context, rules, and feedback loops that it needs less supervision each time.

This has saved me a surprising amount of time, tokens, and rework.

[Priya Roul @priyaaroul]

@shivanshgupta36 @animeshsingh38 BLR ke progressive roads

[Priya Roul @priyaaroul]

stuck here for past 30min, day by day blr traffic is just getting worst.