Nishanth Basava

Skills

Languages

Python

C++

TypeScript

SQL

R

Java

ML / Data

PyTorch

Hugging FacePEFT/LoRA

scikit-learn

NumPy

PandasRAGLLM evaluation

Backend / Infra

FastAPI

Node.js

PostgreSQL

Redis

Docker

AWS (EC2, S3)

SLURM/HPC

Git

GitHub Actions

Frontend

React

Next.js

Vite

Tailwind CSS

Experience

Software Engineering InternIncoming

Fall 2026

Frizzle (YC S25)

Machine Learning Research Intern

May 2026 – Present

Neurdy Lab @ VALIANT · Vanderbilt University

Built a Python fMRIPrep → nuisance-regression → ROI-connectivity pipeline for a 413-subject (458-scan) ADNI resting-state fMRI cohort, parallelized with a 458-task SLURM job array on Vanderbilt's ACCRE cluster.
Developed AutoQA's review layer: a React + FastAPI + PostgreSQL app with role-based access, JWT auth, an append-only audit trail, and per-route latency instrumentation — containerized with Docker Compose for lab infrastructure.
Found a session-pooling bug that flipped 22/458 QC labels and a ~20× scanner-vendor bias in standard outlier criteria.

PythonSLURM/HPCReactFastAPIPostgreSQLDockerLab Website

Data Science Intern

Sep 2023 – Aug 2025

Harvard Biostatistics / Dana-Farber · Harvard T.H. Chan School of Public Health

Co-authored bayesNMF, an open-source R package with a Gibbs sampler for Bayesian Poisson NMF, recovering mutational signatures across 34 cancer types — published in the Journal of Computational and Graphical Statistics (2026).
Built SigServe, a REST API and job queue around bayesNMF so users can run signature discovery without touching R (see Projects below).

RBayesian MLFastAPIRedisAWS EC2Publication

Projects

AutoQA

Fine-tuned VLM pipeline for fMRI quality control

Built a pip-installable QC pipeline and CLI for 458 ADNI fMRI scans, combining fMRIPrep motion/artifact metrics with RAG-grounded LLM review of 1,374 QC images — identified failures missed by motion metrics and prioritized 19–24% of scans for manual review, with GitHub Actions CI and 74% test coverage.
Curated an expert-labeled fMRI QC dataset (293 scans, 879 images; weighted Cohen's κ = 0.92) and distilled a RAG-grounded Claude teacher into Qwen2.5-VL-7B using LoRA, improving subject-held-out macro-F1 from 79.6% to 84.1% at 82% lower inference cost.

PythonPyTorchFastAPIDockerLinuxGitHub

DiffEx

Evidence-based differential diagnosis engine

Built an explainable differential-diagnosis engine using demographic-adjusted priors and a likelihood-ratio knowledge graph spanning 86 conditions and 455 clinical associations — ranked diagnoses with ICD-10 codes in ~5 ms.
Achieved 61%/78% top-1/top-3 diagnostic accuracy on 23 published clinical vignettes vs. 22% for a prevalence-only baseline; 55-test suite gates merges via CI with continuous deployment on Vercel.
Won the $3,000 Vanderbilt Hult Prize, advanced to the USA East Finals, and piloted DiffEx with six emergency medicine physicians at a Guatemala clinic serving 250+ patients.

Winning the Vanderbilt Hult Prize with DiffEx

TypeScriptReactSupabasePostgreSQLGitHub

SigServe

REST API and job queue for Bayesian mutational-signature discovery

Wrapped the bayesNMF sampler (via rpy2) in a FastAPI service with Redis/RQ workers, so users submit mutation-count matrices and poll results without ever running R.
Asynchronous job execution with a PostgreSQL job store, API-key auth, persistent job state, and structured failure reporting; 38 unit/integration tests (88% coverage) with GitHub Actions CI gating merges — Dockerized and deployed on AWS EC2.

FastAPIRedis/RQrpy2PostgreSQLDockerAWS EC2GitHub

Publications

bayesNMF: Fast Bayesian Poisson NMF with Automatically Learned Rank Applied to Mutational Signatures

Jenna M. Landy, Nishanth Basava, Giovanni Parmigiani

Journal of Computational and Graphical Statistics · 2026

hey, i'm

About

Vanderbilt University