Pradyun Kotcherelakota
Skip to the workResume

Avyakt Pradyun Kotcherelakota

Fixing the bug is the easy part. Noticing it is the job.

B.S. Artificial Intelligence at UC San Diego, class of 2029. Backend engineer intern at SkyAccess, researcher in two UCSD labs, and the builder of Research Radar, a live search engine over $1.48B of university research funding.

Open to Fall and Summer software and ML internships

akotcherelakota@ucsd.eduResumeAvyaktKotain/pradyunk

B.S. Artificial Intelligence·UC San Diego·Expected June 2028·La Jolla, CA

Work

Eight pieces of work. Open any of them.

Zero output mismatches

Every output of the ported model was checked against the original. Zero mismatches, out of every row compared. It became the company's first production ML model, and the template for every handoff since.

0
output mismatches
1st
model live in production

Every UC San Diego professor, and who is funded right now

One blip per professor, 2,732 of them, on a radar scope. The bright ones hold an active NIH or NSF grant on record. The rest are not unfunded, much research runs on money the federal databases never see, and the site says so instead of leaving a gap.

2,732
faculty, one point each
585
with an active federal grant on record
$1.48B
linked to its source record

Live figures from researchradar.xyz, refreshed hourly.

346 genes that mattered, out of 8,760

In brain tumour cells that resist treatment, 8,760 genes were measured and 346 showed a real difference. The strongest signal sat in the machinery controlling how cells divide. Every dot here is one of those genes.

346
genes with a real difference
10
biological processes flagged

Experience

Since Jun 2026UC San Diego

Undergraduate Research Assistant · CV Lab, UCSD

InteriorAgent, LLM program synthesis for 3D layout from UCSD and Qualcomm: code, test scenes, and merged fixes. Also wrote the labelling standards for a desktop-use agent that drives Unreal Engine.

Since Jan 2026UC San Diego

Undergraduate Researcher · Rana Lab, UCSD

Which genes behave differently in treatment-resistant brain tumour cells, and an analysis pipeline whose results always trace back to the run that produced them.

Projects

Research Radar

Aug 2026

Shows which UC San Diego professors are funded and publishing right now, with a source link behind every claim.

A public search index of 2,732 UC San Diego faculty, linking $1.48B in active NIH and NSF funding and 3,081 arXiv papers to their source records. Underneath is an identity-resolution pipeline that decides whether a grant belongs to a person across four sources that disagree about names, and refuses to guess when it cannot tell. Launched after a five-lens audit, every fix pinned by a mutation-verified test.

2,732
faculty indexed
$1.48B
grant funding linked to its source
3,081
papers linked to their source

Live figures from researchradar.xyz, refreshed hourly.

PythonFastAPISQLAlchemyPydanticPostgreSQLReactRailwayVercel
Open the live site

Brain MRI Tumor Classifier

Sep 2025

Reads a brain scan, says which of four tumour types it is, and shows you why.

92% accurate across four tumour types. Every answer comes with a picture of the regions of the scan that led to it, so a doctor can check the reasoning instead of taking it on trust. Deployed as a web app anyone can try.

Best Project, UCSD Summer Program for Incoming Students (audience of 50+)

92%
classification accuracy
4
tumour types classified
PythonTensorFlow/KerasLIMEGradioscikit-learn
Source

Toolkit

Languages

  • Python
  • TypeScript
  • JavaScript
  • SQL (Postgres)
  • Java
  • C/C++
  • R
  • HTML/CSS

Frameworks

  • Node.js
  • React
  • FastAPI
  • SQLAlchemy
  • Pydantic
  • Flask
  • Next.js
  • Prisma

ML & Data

  • PyTorch
  • TensorFlow/Keras
  • scikit-learn
  • ONNX Runtime
  • pandas
  • NumPy
  • Matplotlib

Tools

  • Git
  • Docker
  • Jest
  • pgvector
  • Jupyter

Bioinformatics

  • BWA-MEM
  • CIRI3
  • goatools
  • GSEA
  • SRA Toolkit

Highlights

  1. Shipped a company's first production ML model, as an intern.

    Ported a data scientist's model into the live stack and proved it identical to the original: zero mismatches. It became the template for every handoff since.

    1stmodel a company ran for real

  2. Built and launched a live product, end to end.

    Research Radar: pipelines over four public data sources, an identity-resolution engine, an API, a React front end, and the deploy. 2,732 professors and $1.48B of funding, searchable today.

    $1.48Bresearch money made searchable

  3. Finds the failures that raise no alarm.

    A bug quietly erasing grants, a database running the wrong function, an AI building impossible rooms. None of them errored. All of them were found and fixed.

    0alarms it ever raised

  4. Researches on two fronts while a full-time student.

    Code and test scenes for a 3DV 2026 paper's codebase, from UCSD and Qualcomm, and a gene-expression pipeline in a cancer lab.

    0warnings it ever gave

  5. Proves it with tests, then attacks the tests.

    More than 1,500 automated tests behind Research Radar, and every launch fix pinned by a test that was itself checked by mutation.

    10×faster under a crowd

  6. Judged best in the room.

    Best Project at UC San Diego's Summer Program for Incoming Students, for a tumour classifier that shows its reasoning: 92% accuracy across four tumour types.

    92%accuracy across four tumour types