Khush Shah
Carnegie Mellon University | B.S. in Computer Science & Mathematics | Expected May 2029
Fully work authorized | U.S. Citizen
Education
B.S. in Computer Science & Mathematics
Carnegie Mellon University
Expected May 2029
- Relevant classes: Machine Learning, Data Structures & Algorithms, Competition Programming, Computer Systems, Linear Algebra, Multivariable Calculus, Differential Equations.
- Focus areas: Machine Learning, Artificial Intelligence, Backend Systems, Data Structures & Algorithms.
High School Diploma
Los Altos High School
Technical Skills
Programming Languages
Languages
- Python, Java, C++, C, C#, JavaScript, TypeScript, Rust.
Developer Toolkit
Tools & Frameworks
- PyTorch, NumPy, Pandas, React, Node.js, Flask, Express, Postgres, ORM, Unity, Git/GitHub.
Machine Learning
ML
- Implemented transformers from scratch; completed ARENA, a technical AI safety curriculum covering reinforcement learning, evaluations, interpretability, and alignment.
- Built working knowledge of LLM internals, attention, RLHF, and alignment techniques.
Projects & Research
Carnegie AI Safety Initiative
CASI
2025 - Present
- Participated in a technical AI safety research reading group, studying fundamentals of ML papers and alignment techniques including transformers in depth, mechanistic interpretability, and RLHF.
- Co-leading a 3-person research project on AI safety alignment robustness, producing empirical research.
Geometric Framework for Predicting Fragility of LLM Safety Training Methods Research Paper
- Used LAT-extracted RepE vectors and Hessian eigenvector analysis to show safety concepts occupy lower-rank, higher-curvature subspaces than capability directions.
- Showed benign fine-tuning preferentially aligns with fragile safety directions, with representational rank and layer depth as independent fragility predictors.
- Built a predictive framework ranking DPO versus PPO-RLHF alignment robustness by geometric properties, validated on HarmBench across LLaMA, Mistral, and Qwen.
CMU Courses Backend
ScottyLabs Tech Team
2025 - Present
- Built automated Rust system to query CMU APIs and recursively discover related course resources.
- Wrote Python ETL to parse and normalize data and maintain Postgres schema and records.
- Exposed processed datasets to the frontend through structured JSON and CSV endpoints.
- Rebuilt the web app in TypeScript and React using TanStack Router; platform used by 100s of CMU students.
Stock Price Forecasting Model
Team Lead - UC Davis COSMOS
July 2024 - August 2024
- Implemented ML models from scratch in Python using only NumPy, deriving forward and backpropagation manually from calculus and linear algebra primitives.
- Led 5-person team supervised-fine-tuning sentiment-analysis models for stock-price forecasting; outperformed baselines and presented results at COSMOS Showcase.
Leadership
Organizer & Sponsorship Lead
Los Altos Hacks
2021 - 2025
- Organized the world's largest high-school hackathon and raised $15K+ in sponsorships.
- Gave speeches to 1000+ students and mentored many new members.
Senior VP
Computer Engineers of the Next Generation (CENG)
2022 - 2025
- Managed 100+ volunteers teaching 24+ coding classes in Python, Java, and JavaScript to underserved schools.
- Earned the Presidential Volunteer Service Award for 200+ service hours in a single year.