Hold on
This one's desktop-only for the moment — mobile's still being built. Pull it up on a laptop or desktop (900px and wider) to actually see it.
Index
Retrieval, pipelines, and the parts that decide when not to answer. Five of them.
AI & ML, third year, Hyderabad.
Work
Five systems. Each one built around the thing it refuses to do.
Answers from what the index actually returned, or from nothing.
Turns a map search into contactable leads, or leaves the contact field blank.
The model decides what each scene shows. Real audio decides when it shows it.
Turns a meeting transcript into owned, dated tasks. Never invents an owner.
Screens recruiter mail for fraud, and says so when it can't tell.
Work
Answers from what the index actually returned, or from nothing.
Turns a map search into contactable leads, or leaves the contact field blank.
The model decides what each scene shows. Real audio decides when it shows it.
Turns a meeting transcript into owned, dated tasks. Never invents an owner.
Screens recruiter mail for fraud, and says so when it can't tell.
Method
Skipping one doesn't save time. It just moves the cost later.
Read what already exists — the documents, the logs, the thing that broke — before writing anything.
Did it look in the wrong place, or think about it wrongly? Those need different fixes.
Remove before adding. Most problems get solved by deleting a step, not writing one.
Say exactly what goes in, what comes out, and what happens when it fails.
Test the same questions every time. A result you can't repeat isn't a result.
Method
Skipping one doesn't save time. It just moves the cost later.
Read what already exists before writing anything. Skip it and you solve a problem nobody had.
Did it look in the wrong place, or think about it wrongly? Skip it and you tune the model when the retrieval was wrong.
Remove before adding. Skip it and you automate something that should have been deleted.
Say exactly what goes in, what comes out, and what happens when it fails. Skip it and the failure mode gets discovered by a user.
Test the same questions every time. Skip it and every improvement is a story.
Picked for the job, not the résumé.
Stack · 02
Most of these are common. Three of them are not.
About
Third year of a B.E. in Artificial Intelligence and Machine Learning at Lords Institute of Engineering and Technology, Hyderabad. Graduating 2028.
About
I started building systems in first year instead of waiting for the degree to finish. Three of the five above came out of that.
The other two were built at Oyelabs, on an AI Automation internship that's still running. Requirements arrive from somebody else there — a different discipline from choosing your own.
The degree provides the scaffolding. The projects build the muscle.
Most of it was written after midnight. Not discipline — that's just when the room goes quiet.
Web development is the hobby that got out of hand. This page is hand-built: every section, every transition, no template. It's where an idea gets tested against a browser instead of a notebook.
Away from the screen: chess, cricket, film, and whatever's playing.
Contact
Retrieval that has to be right. Agents that have to fire once. Generation that has to stay inside a schema.
Open to internships and project work. More interested in the problem than the title.
Writing about a corpus? Say what's in it, what questions it needs to answer, and what a wrong answer would cost. That saves a round trip.