Research Interests
My research interests span network protocol design, distributed systems, and applied cryptography, with a focus on how security and correctness properties can be designed into networked systems from first principles. I approach this from both directions: building protocols that make strong security guarantees in practice (TYPHOON, a privacy-preserving UDP transport; Airwire, a hybrid web/SMS messenger) and reasoning formally about correctness under adversarial and uncertain network conditions. For doctoral research, I am interested in exploring this boundary more broadly — across wireless and IoT protocols, distributed consensus and fault tolerance, robotics and autonomous systems, and security analysis at the network and transport layers. I aim to work at the intersection of rigorous theoretical foundations and real-world protocol engineering.
Keywords: Network protocol design; Distributed systems; Peer-to-peer networks; Applied cryptography; Transport protocols; Wireless and IoT security; Distributed algorithms; Robotics and autonomous systems; Peer-to-peer networks; Traffic analysis; Formal protocol specification
Papers
Specifies TYPHOON, a UDP-based transport-layer protocol designed to make traffic identification impractical. All packets are fully encrypted with no cleartext fields; decoy traffic and fake protocol headers mimic generic network behavior. The protocol uses a session manager / flow manager architecture, supports multiple proxy endpoints, and offers post-quantum readiness via Classic McEliece KEM alongside XChaCha20-Poly1305 and AES-GCM-256 symmetric modes. Includes a Rust reference implementation.
Technical Reports
Compared two person detection methods for 2D LiDAR data: a heuristic cluster-based algorithmic detector and a deep learning DROW detector based on 1D convolutional architecture. Developed the Follow-the-DROW (FTD) library, ROS package, and Docker image for detector comparison and deployment on the RobAIR platform. Identified significant annotation quality issues in the DROW dataset affecting evaluation reliability; full quantitative comparison was not completed due to computing resource constraints.
Theses
Designed and implemented a schema-driven code generation system for Ansys's simulation framework (DPF), replacing manually maintained client libraries in Python, C++, and C# with auto-generated code from a single schema definition. Evaluated multiple schema technologies (FlatBuffers, Protobuf, Cap'n Proto, etc.), designed a front-end/back-end generation pipeline with intermediate representation, and integrated the generator into existing CMake and Poetry build systems.
Developed a monitoring system for Sosnowsky's hogweed (Heracleum sosnowskyi), an invasive phototoxic plant species. The system combines a Django/PostGIS server backend, a Flutter/BLoC mobile client for field reporting, and a TensorFlow-based ML image classification module for species identification. The classifier uses MobileNetV2 transfer learning (pretrained on ImageNet) to distinguish three classes (hogweed, cetera, other) over a dataset of 7100 images per class sourced from OpenImages and iNaturalist, achieving 92% accuracy with 0.4s inference time and a 79.3 MB TensorFlow Lite model. The backend uses PostGIS/PostgreSQL for geospatial tracking of reported sightings.