Search systems, geocoding, place visit detection, sensor fusion, machine learning

Worked primarily on HorizonDB, Radar's proprietary geocoder, written in Rust.

Developed algorithmic optimizations that improved search performance by up to 99% in some cases, reducing latency from hundreds of milliseconds to sub-millisecond ranges.

Built data pipelines using Apache Spark and implemented net-new search functionality for street-level geocoding, enabling Radar to unlock significant growth opportunities in international markets.

Consulted on firmware development for Radar's new indoor geolocation solution, designing mathematical algorithms for signal processing and location estimation in Python.

Worked closely with the GTM team to design statistical tests demonstrating the effectiveness of Radar’s products to clients, providing best practices for measuring efficacy and showcasing improvements gained from switching geocoder providers.

Undergraduate Student Researcher in the Applied Math department.

Worked with N. Sri Namachchivaya on non-parametric quickest-change detection methods. Investigated geometric methods for detecting changes in unknown probability distributions in an online regime.

Worked with David Del Rey Fernandez and Roberto Guglielmi on investigations regarding the use of hardware Spiking Neural Networks (SNNs) for solving PDEs. Hardware SNNs have the potential to be used as low-power accelerators for conventional computers, particularly for Monte Carlo simulations of complex phenomena.

Performed research on localization methods for Bluetooth Low Energy (BLE) emitters in dense indoor sensor networks. Research involved probability theory, convex optimization methods, estimation methods (via Kalman filtering and particle filtering), as well as machine learning using TensorFlow.

Designed and implemented machine learning-based fingerprinting method for localization in noisy environments.

Made improvements to real-time location services API. Developed and integrated a more robust Kalman filter using Python, MongoDB, and Redis.

Contributed to the development of MATLAB Wireless Testbench, a tool facilitating rapid prototyping of software-defined radio systems.

Implemented process improvements that reduced the number of FPGA bitstreams needed for building and testing Wireless Testbench by 50%, resulting in significant time (several hours) and computational resource savings.

Enhanced the high-speed data acquisition interface (C++) for accurate performance benchmarking, ensuring optimal system evaluation.

Made valuable improvements to the open-source UHD simulator (Python), enabling cost-effective testing without expensive hardware. These enhancements improved the simulator's versatility, allowing for dynamic re-routing of RFNoC radio signal processing blocks during tests.