Researcher · Builder · Harvard
Cristiana Murgoci
Researcher at the intersection of AI safety, statistics, and physical automation.
Bachelor's in Statistics & CS · Harvard College
Master's in Statistics · Harvard GSAS
AI Visual Discovery
Building and open-sourcing vision-language pipelines that generate catalog metadata for special-collections photographs, working closely with librarians to automate the description and digitization of Houghton Library's collections at scale, and evaluating the pipelines against expert records so the methods stay accurate, reproducible, and open. Part of "Beyond the Frame," a Harvard Library Advancing Open Knowledge grant.
Market Making for AI Alignment ↗
Proposes training a market-maker model to forecast a human's reflective judgment after exposure to all relevant arguments, while adversaries surface information that most shifts that forecast. Developed during the AISES fellowship (Summer 2025). A simulation probe showed that prompting alone makes the market-maker worse because it over-corrects on weak counterarguments, establishing why training competitiveness is the core research question. The full proposal pairs MM with ELK-style probes, process supervision, and cross-examination, evaluated against RLHF and Debate baselines on truthfulness, calibration, and deception robustness.
Characterizing Collusive Dynamics in Multi-Agent LLM Systems ↗
Designing and evaluating multi-agent simulations where LLM agents act in market environments, with Prof. Shi Feng (GWU) and Arush Tagade. Key finding: agents collude significantly more when they can communicate and recognize shared model weights. Building tools to measure and mitigate systemic risks from emergent agent coordination.
AI Safety Policy Fellowship
Selective fellowship on AI governance and safety. Covered technical foundations of machine learning, risks from advanced AI, compute governance, corporate responsibility, and international policy frameworks. Engaged with research, government reports, and regulatory proposals on frontier AI systems and safety standards.
AI Safety Technical Fellowship
Analyzed reinforcement learning from human feedback (RLHF), goal misgeneralization, mechanistic interpretability, and red teaming for AI systems. Developed skills in AI model evaluation, deceptive behavior analysis, and safety regulations for industrial-scale AI, contributing to discussions on policy and alignment strategies.
Citadel Securities Quant Invitational
Designed and implemented an ETF arbitrage bot capitalizing on market inefficiencies. Developed a statistical arbitrage strategy trading pairs of correlated stocks within a six-asset portfolio, achieving a Sharpe ratio of 55.8 in a simulated random walk environment.
Insight Program
Highly selective program introducing undergraduate students to market-making strategies including arbitrage, through trading simulation games.
First-Year Trading and Technology Program
Highly selective program introducing first-year undergraduates to Jane Street's trading and technology models. Participated in classes and team-based mock trading simulations.
Quantum ML for Neutrino Detection
Created a quantum data processing protocol for the IceCube neutrino experiment using quantum annealing, in the Carlos Argüelles Group. Wrote an optimization algorithm achieving exponential speedup over classical methods for neutrino event analysis. Work presented at Fermilab.
US Quantum Information Science Summer School
Intensive summer school on quantum computing at Fermilab. Track: Qubits, Simulation, Quantum Software, Algorithms, and Applications. Studied combinatorial optimization, QAOA, quantum error correction, and QPUs. Implemented quantum algorithms in Qiskit and QuTiP.
Research Assistant
Research on applications of electromagnetic forces in medicine, in the Dr. Eng. Mircea Ignat Group. Investigated using flexible electrodes and magnetic fields to treat atherosclerosis, nanometric carbon particles for faster drug delivery, and magnetic field modulation of blood viscosity as a non-invasive alternative to aspirin.
Physics Unlimited Explorer Research Competition
Led teams across three years in a research competition organized by Princeton alumni. Projects: path integral approach to quantum mechanics and Feynman diagrams (Honorable Mention, 2022), designing a quantum cascade laser (Honorable Mention, 2021), and modeling orbital resonance (Special Award, 2020).