Charbel Chávez

• Designing and implementing a hybrid classical-quantum architecture for a benchmark, integrating classical embeddings with variational quantum layers.
• Building reproducible experimentation pipelines for NLP tasks (IMDB, AG News) and computer vision tasks (MNIST, Fashion-MNIST).
• Automating experiment execution and managing version control for the team's shared repository.
• Collaborating on a multifactorial experimental protocol evaluating qubit count, encoding strategy, and circuit depth.

AI Systems & LLMs

Built an end-to-end pipeline for HackPrinceton 2026 transforming user drawings into structured JSON representations, enabling real-time game reconstruction. Integrated Gemini API for multimodal reasoning to interpret drawings and generate feedback on playability, difficulty, and level design quality. Framed level design as an AI-driven iterative refinement problem, earning Best Rookie Track and 1st Runner-Up for Best Use of AI + Hardware.

Quantum Machine Learning

Designed a 2-qubit parameterized quantum circuit with RY-RZ-CNOT ansatz trained via the parameter-shift rule. Benchmarked against SVM-RBF baseline, achieving 91.2% test accuracy with near-identical AUC (0.9868 vs 0.9871). Analyzed loss landscape, quantum kernel geometry, and barren plateau theory to explain the performance gap.

Full-Stack & Android Developer

Full-stack healthcare platform to manage patient appointments and electronic medical records for a nephrology organization. Built RESTful backend services with Node.js and Express, a React web app for doctors, and an Android application in Kotlin enabling patients to book appointments and access lab results.

Machine Learning Engineer

Trained and evaluated LightGBM models on 15,000+ Kepler and TESS samples for exoplanet detection. Built a FastAPI backend and React frontend for real-time inference and visualization. Achieved 2nd place at NASA Space Apps Challenge 2025 (local event).

Agent-Based Modeling

Developed a multi-agent system using Python and Mesa, achieving a 27% win rate across 100+ simulations. Implemented autonomous agents with dynamic role assignment and Dijkstra-based pathfinding. Built real-time visualization using Flask and WebSockets.