Skip to content

I'm a builder who owns problems end-to-end and ships things people use.

Currently…

  • Team Lead @ Cornell AppDev
  • TA, CS 4820 @ Cornell
  • CS '27 @ Cornell Engineering

Next…

  • Member of Technical Staff Intern @ Forus
  • Jan 2027 · New York

Previously…

  • Datadog
  • Amazon Web Services (×2)

LATEST · Nov 18, 2026Wrapped my Datadog internship. Joining Forus in NYC this January.

Selected work

6 case studies

Forus wordmark

Forus: Getting medicine into people's hands faster

Starting Jan 2027 | Member of Technical Staff Intern · New York

Start
Jan '27
Location
NYC
Intern
MTS
Stefanie pointing to the Datadog logo on a purple wall at the Datadog office

Datadog: An LLM pipeline that translates cloud security rules, taken from 30% to 70% accuracy

Summer 2026 | SWE Intern · K9 Cloud Security Misconfigurations

Translation accuracy
30→70%
Eval checks per rule
4
Shipped via GitLab
CI
Cornell AppDev members posed on a staircase holding an AppDev sign

Cornell AppDev: Leading a 71-person team that ships 7 apps to 14,000+ students

Jan 2026 – present | Team Lead · elected by a 71-member org

Students served
14k+
Production apps
7
Members led
71
Stefanie speaking on stage with a microphone at an AWS Intern Day event, with the audience behind her

Amazon Web Services: Laying IAM foundations for trillions of authorization requests a day

Summers 2024 & 2025 | SDE Intern ×2 · Identity and Access Management

Auth requests / day
1T+
Summers on IAM
2
+ Java
Rust
Uplift app icon beside two phone screens: a gym list with live capacity, and a profile with workout goals

Uplift: Cornell's all-in-one fitness app, from gym capacity to workout tracking

Live since 2018 | Cornell AppDev · iOS

Live since
2018
Platform
iOS
Score app icon beside two phone screens: upcoming games, and a Cornell vs. Yale box score

Score: The sports hub for Cornell Athletics, launched Spring 2025

Launched Spring 2025 | Cornell AppDev · iOS

Public launch
SP '25
Scores + stats
Live

Research

In progress · Fall 2026

Performance Analysis of Fused MoE Inference in vLLM on 4×A100 GPUs

Timothy Adeyemi, Temi Adebowale, Abrar Amin, and Stefanie Rivera-Osorio

Mixture-of-Experts models run only a few experts per token, but split that work unevenly across GPUs. We profile how vLLM's fused MoE kernels spend their time serving Qwen3-30B-A3B on four A100s, measuring expert load per GPU and per layer, and how traffic patterns trade throughput for latency.

Avg per-layer imbalance
1.39×
Busiest GPU, worst layer
+71%
Throughput under burst
6×

Materials: vLLM · Qwen3-30B-A3B · Triton · 4×A100 · Perlmutter

Open full poster (PDF) ↗
Research poster: per-GPU kernel time breakdown, per-layer routing imbalance chart, and serving performance table for fused MoE inference in vLLM
Research poster · click to open the full PDF

Side projects

  • Road English Coach · Big Red Hacks 2026

    A voice-first coach that quizzes drivers on the road signs they just passed. Sign recognition runs on-device, and ElevenLabs handles speech.

    Python · YOLOv8 · Apple Vision · ElevenLabs

  • FootySearch · Cornell INFO 4300

    An information retrieval system for soccer players. I designed and built the frontend.

    React · Figma

Experience

Details →
  1. Member of Technical Staff Intern @ ForusNEXT

    Starting Jan 2027

  2. Team Lead @ Cornell AppDevNOW

    Jan 2026 – present

  3. Teaching Assistant, CS 4820 @ Cornell Bowers CISNOW

    Jan 2026 – present

  4. Software Engineering Intern @ Datadog

    May – Aug 2026

  5. SDE Intern, IAM @ Amazon Web Services

    May – Aug 2025

  6. SDE Intern, IAM @ Amazon Web Services

    May – Aug 2024