Aarav Sharma · Portfolio

About

I enjoy building products that feel simple on the surface and rigorous underneath. Recently, I have worked across mobile, applied ML, and research: shipping user facing features at Amazon, building vulnerability detection pipelines, and modeling fluid flow in cross flow heat exchangers.

Experience

Amazon

Software Development Engineer Intern

May 2025 to Aug 2025 · Seattle, WA

Engineered a React Native carousel for Echo devices, increasing one tap photo additions by 36 percent for about 15.5M MAU.
Built a comprehensive Jest suite with 95 percent coverage across 4 user flows.
Led feature gated rollout and success metrics, driving a 28 percent engagement lift.

ManTech

Software Engineer Intern

May 2024 to Apr 2025 · Herndon, VA

Trained Graph Neural Networks to identify software vulnerabilities in binaries, more than 90 percent accuracy.
Developed a custom LLM in SGLang for commit analysis and CWE classification; plus 25 percent threat detection.
Automated analysis and triage of 60 plus vulnerabilities; reduced threats by 30 percent.

Georgia Tech DICE Lab

Undergraduate Research Assistant

Dec 2023 to Present · Atlanta, GA

Computational analysis of a cross flow heat exchanger using topology optimization.
Enhanced location detection with CNNs in PyTorch.
Processed 15 plus CFD features from OpenFOAM to predict fluid flow phenomena.

Projects

Spotify Wrapped App with Personalized LLM Analysis

Android app with dynamic music visualizations and LLM powered insights for 150 plus users.

Implemented OAuth2 with Firebase Authentication and Google Sign In with sub 200 ms median auth.
Built an LLM summary pipeline to generate per user insights; cached results to reduce tokens by 60 percent.
Shipped Crashlytics and analytics to iterate on UX based on retention and session length.

Full Stack College Scheduler App

Android plus Firebase app helping 500 plus students manage schedules with conflict detection.

Designed a conflict detection algorithm for overlapping courses; reduced schedule errors by about 40 percent in tests.
Integrated push notifications with FCM; built a rules engine to notify on conflicts and upcoming exams.
Provisioned Firebase security rules and batched writes to keep operations under 200 ms P50.

Statistical Analysis and Evaluation of Tennis Games Using Computer Vision

TrackNet plus OpenCV to detect bounce locations with more than 95 percent accuracy. Visualized 100 plus rallies; presented findings to 300 plus attendees.

Reproduced TrackNet inference and added post processing to estimate velocity and bounce confidence.
Exported rally segments to CSV for downstream analytics; created overlays for highlight reels.
Deployed a lightweight Flask API to serve predictions for a demo front end.

Skills

Python Java C/C++ JavaScript TypeScript Swift Kotlin SQL PostgreSQL MongoDB HTML/CSS React React Native Node.js Express Django Flask PyTorch TensorFlow scikit-learn OpenCV SGLang Docker Kubernetes AWS Firebase Git Linux Bash