Survey Selection and the Origin of Cold Rocky Super-Earths

ETH Zürich, May 2026–Present · Supervisors: Prof. Caroline Dorn, Annika Salmi

Python, NumPy, SciPy, pandas, AstroPy, Matplotlib, P-Pop

• Built detection models for Kepler, TESS and precision radial velocity (HARPS/NIRPS), tested against official pipelines on 7,995 KOIs, 7,307 TOIs and 2,510 published RV semi-amplitudes.
• Designed a prior-free population generator for volume-limited 60 pc Gaia stellar samples: under 1 s per 10⁶ planets, replacing a ~6-hour P-pop run.
• Ran two formation scenarios through the detection model and compared posterior-predictive volatile fractions against the NASA Exoplanet Archive over 5,000 Monte Carlo draws.

Protest Event Data (ROP299)

Munk School of Global Affairs, May–Aug 2025 · Supervisor: Prof. Lynette Ong

• Manually coded 500+ protest cases into a team dataset of 7,000+ categorical records from five news/NGO sources.
• Analyzed monthly trends and provincial distribution from 2023–2026, focusing on housing protests and migrant-labor cases; over a quarter of cases came from Guangdong and Henan.

Social Activity and Depression (PHQ-9)

University of Toronto, Jan–Feb 2025 · Supervisors: Prof. Scott Schwartz & Dr. Kiffer Card

• Built OLS and logistic-regression models predicting PHQ-9 depression scores from social activity, comparing in-person and online interaction.
• In-person interaction outperformed digital connection for well-being (p < 0.001).

Can Super-Earths Be Born Rocky? Separating Survey Selection from Formation in the Deficit of Large Rocky Planets at Low Insolation

A. Salmi*, H. Mei*, C. Dorn · In preparation (2026) · *Equal contribution

Passion Finder: AI-Powered UofT Course Explorer

Nov 2025 · Java

• LLM-powered course recommender built for a course final project. Wrote 35% of the codebase in Clean Architecture and the API integration in a 5-person team.

Neural Networks from Primitives

Sep 2025 · PyTorch

• Implemented and trained a CNN (MNIST), a recurrent network and a multilayer perceptron from PyTorch primitives, covering backpropagation, optimization and regularization.

Reinforcement Learning Playground

Sep 2025 · Python

• Monte Carlo state-value estimation for a stochastic grid-world (FrozenLake), comparing first-visit value estimates against episode count.

CaYPT: Oscillating Sphere

Mar 2023 · Team Captain, Silver Medal

• Captained a team of 5; built a Helmholtz coil pair and the full apparatus, fitting results to theory within 6.5%.