Aarya
Srinivasan

I'm a student at Berkeley working on machine learning and AI. Right now I'm a founding engineer at NextXI, where we turn cricket footage into coaching metrics that help young players improve and get noticed. I've also spent a lot of time on LLM systems, cutting inference costs and making AI more reliable.

UC BERKELEY ’28/BS COMPUTER SCIENCE · BA DATA SCIENCE
RÉSUMÉ →
HOVER TO RESOLVE
PYTHON · ASYNCIO · FASTAPI · C/C++ · SQL · R · JAVA · JAVASCRIPT · TYPESCRIPT · NODE · REACT · NEXT.JS · PRISMA · PYTORCH · TRANSFORMERS · HUGGING FACE · CUDA · TRITON · FLASHATTENTION · VLLM · SGLANG · RAG · YOLO · OPENCV · SCIKIT-LEARN · KALMAN FILTERING · LLMLINGUA · DOCKER · AWS SQS · POSTGRESQL · SUPABASE · GIT · LINUX ·

LLMBDA

The decision layer for AI inference
75% cheaper
with better performance
READ THE WRITE-UP →

NextXI

AI-backed scouting for young athletes
250+ players
and 10+ coaches across real academies
READ THE WRITE-UP →

Sid

Enabling determinism in LLM inference
Bit-identical
the same prompt returns the same bytes across all runs
READ THE WRITE-UP →

Experience

NextXI
FOUNDING ENGINEER

Trained custom YOLO detectors alongside a pose estimation model to track body keypoints, then wrote the OpenCV post-processing pipeline with Kalman filtering that stabilizes the output into clean joint trajectories and derives coaching metrics. Shipped the production Next.js / TypeScript / Prisma scouting platform, connected to the CV service by a Postgres-as-a-queue worker with idempotent job handling.

2026 —
UC Berkeley
DATA SCIENCE RESEARCH ASSISTANT

Processed raw 10-K filings into corporate restructuring and reorganization datasets, implemented the validation protocols that resolve noise for downstream econometric and ML modeling, and programmed a Hugging Face and LLM pipeline to automate scraping and processing of SEC filing data.

2026
AshaDx
SOFTWARE ENGINEERING INTERN

Researched ML architectures for an AI pipeline automating de novo antibody design for clinicians. Designed and built the company database in Supabase, and automated contact routing with Make webhooks, eliminating manual lead triage.

2026
Sync Visual Media
SOFTWARE ENGINEERING INTERN

Built full-stack React and Python client platforms, lifting visitor-to-customer conversion roughly 15% per client, and engineered SEO rendering pipelines (metadata tagging, structured schema, image indexing) across 100+ digital assets.

2025
UQ Child Health
RESEARCHER & DATA ANALYST

Analyzed 500+ pediatric intensive care studies into a standardized outcome-set database, designed the REDCap collection system, and authored a systematic review protocol registered on PROSPERO.

2024

Footnotes

Other work, projects, coursework, awards, and education.
OPEN →