A distributed backtesting prototype built for CMU’s Distributed Systems coursework. It separates metadata coordination, job execution, and chunked market-data storage so each has a clear role.
Coordinators use Raft to replicate metadata and elect a leader. Workers retrieve data over RPC, execute jobs, and return results—exploring how research workloads can run across machines.
A high-performance American option pricing project for CMU 15-418. The solver simulates asset-price paths and uses least-squares regression to estimate whether an option should be exercised early or held.
The repository explores a baseline implementation alongside optimized CPU, MPI, and CUDA GPU variants, with benchmark scripts and reports for studying parallel execution.
SAMS separates a shared Transformer-based market interpreter from 24 independent, stateful decision agents. Built in PyTorch on Optiver data, it combines their actions through a learned population gate to model future market states.
The public demo explores observed markets and counterfactual rollouts alongside individual agent behavior. The project placed 2nd in the AI Innovation track at the BigQuant 2026 Global Competition North America Final at UC Berkeley.
OUSI brings philosophical thinking into everyday life. It uses AI to help people examine a situation, question their assumptions, and explore different philosophical perspectives.
The product centers on reflection and inquiry: making the ideas and methods of philosophy something people can engage with as they navigate decisions, uncertainty, and the questions that matter to them.
