Systems work outside the research collaborations: low-latency execution, quantitative tooling, and retrieval-augmented agents.
Low-latency systems · market microstructure
A limit order book and matching engine in C++ sustaining roughly three million orders per second. The data structures underneath it are the subject of a reproducible, cross-platform study comparing their behaviour across implementations and hardware.
Stack: C++, lock-free data structures, cross-platform benchmarking
Quantitative research tooling
A backtesting framework in Python and C++ for evaluating trading strategies against historical data, with the performance-critical paths in C++ and the research interface in Python.
Stack: Python, C++
Retrieval-augmented generation · agent tooling
Retrieval-augmented question answering over document collections, and agents that translate natural-language questions into SQL, query the database and summarise what comes back.
Stack: Python, LangChain, vector databases, PostgreSQL