In progress2026
AIKYA
A material-code harmonization engine that refuses to merge anything it cannot justify.
I build AI systems that keep working. And I can show you how I know they work.
I am a Computer Science undergraduate in Delhi, two years into running things for college societies, and I spend most of my own time building applied AI systems. The part I care about is the half nobody puts in a demo: deciding what correct looks like, measuring whether it still is, and refusing to ship the answer that cannot be defended.
Everything below is something I can point at. Skills link to the work that proves them, numbers come with the method behind them, and anything I have not built yet says so.
- Based in
- Gurugram, Delhi NCR
- Studying
- Computer Science, ARSD, University of Delhi
- Skills with evidence
- 17
- Society roles held
- 5
The map
- AIKYA
- Weprax
- Workout Tracker
- ship and groundwork
- Evaluation design
- Python
- Context engineering
- TypeScript
- FastAPI
Projects in the accent colour, skills in blue. The full list of both is below.
Everything I can do
Pick one to see where I used it.
Languages
AI and ML
Backend
Frontend
Practice
The work behind it
- In progress2026
Weprax
Tooling for understanding code you did not write, including the code a model wrote for you.
- Shipped2026
Workout Tracker
A mobile-first PWA for logging gym sessions, in daily use.
- Shipped2026
ship and groundwork
Two Claude Code skills that make a coding agent behave predictably across sessions.
Where I have turned up
- 8.5
- CGPA
- Computer Science, ARSD, University of Delhi
- 2
- Hackathons
- Smart India Hackathon and Paytm Hackathon
- 4
- Languages
- Python, TypeScript, JavaScript, SQL
- 0
- Certifications
- None yet. The slot is built and waiting.
Hackathons
- 2026
Smart India Hackathon
Problem statement SIH26099, Ministry of Petroleum and Natural Gas. AIKYA came out of it.
- 2026
Paytm Hackathon
Certifications, education and society roles are on the background page.
How I work
Decide what correct means first
A model cannot tell you it is wrong, because detecting that needs ground truth it does not have. Somebody has to define the answer before anything is measured against it.
Measure it, do not judge it
Numbers beat impressions, and a number nobody can reproduce is an impression. Every figure on this site names the method that produced it.
Refusing is a feature
A system that declines the case it cannot justify is worth more than one that answers everything. Uncertainty should reach a person, not get rounded away.
| CPCLFAST-1001 | BOLT SS304 M12 X 50mm IS 1367 |
|---|---|
| IOCLFAST-1002 | BOLT SS316, SIZE: M12 X 50mm, IS 1367 |
- Text similarity
- 97%
- Metallurgy
- SS304 / SS316
- Critical attribute
- Yes
Not merged. Metallurgy is a critical attribute, so a conflict vetoes the match whatever the score. SS316 is specified for chloride service. The similar one is the wrong bolt.
Two material codes that any string metric scores at 97%. One is the wrong steel for the job, so the system refuses rather than guesses.
Looking for an applied AI role where the hard part is making the system trustworthy.