I build software that keeps working when the happy path doesn't.
Backend systems · data infrastructure · applied AI · thoughtful products
Computer Science @ University of Southern Maine · graduating December 2026
I'm a compute enthusiast who enjoys engineering things that aren't always visible in a demo, i.e recovery after failure, clean interfaces, reliable data, and tools people actually want to use.
I have worked on medical-imaging ML systems and document workflows, and I like connecting research ideas to software people can run. Lately, I've been spending most of my time on distributed systems reliability, backend engineering, and data pipelines.
Outside the code, I have a soft spot for aviation, learning new niche subjects, and building small tools that make someone's day easier.
Research paper: Flight Trajectory Estimation under Network Delays
What happens when the data arrives late, goes missing, or isn't quite reliable?
I'm exploring how probabilistic state-space models (Extended Kalman Filters) compare with recurrent neural networks (GRUs/LSTMs) when estimating aircraft trajectories from delayed or missing ADS-B telemetry.
The goal is to understand how different approaches handle uncertainty, observation delays, and real-time state estimation.
Python · Time Series · Kalman Filtering · PyTorch · Probabilistic Modeling
Still a work in progress — experimenting, benchmarking, and learning along the way.
|
sync-sdk A Python SDK with durable outbox storage, retry and replay safety, conflict checks, database adapters, and a synchronization gateway.
|
Automated reconciliation engine Multi-provider ingestion, auditable reconciliation, exceptions, DuckDB, dbt, and documented benchmarks (~33k rows/sec in tested workloads).
|
|
StackTrace.run An interactive incident simulator with a mock terminal, real browser-based SQL, authenticated user accounts, and AI-assisted scenario generation.
|
Shipmates A study-buddy platform for course groups, resource sharing, flashcards, quizzes, and AI-assisted study content.
|
|
CNN vs. ViT for brain MRI A comparative deep-learning experiment on MRI images, with CNN and vision-transformer approaches and documented results.
|
Air cargo risk analytics A local ELT analytics workflow with bronze/silver/gold models and BI-friendly outputs for shipment risk analysis.
|
- Reliable by default: failures should be anticipated, observable, and recoverable.
- Readable systems: documentation and developer experience are part of the product.
- Useful AI: models are most interesting when they solve a concrete problem.
- Learning in public: experiments, benchmarks, and trade-offs should be easy to inspect.
Medical imaging ML: worked on mammography-related model serving and developer tooling through Rayomedic–USM.
Product building: contributed document-template and interoperability workflows for Varelio, informed by feedback from credit-union staff.
Teamwork: second place at the 2026 USM Business Analytics Hackathon with an NLP + analytics project built in a short sprint.
Distributed systems, synchronization semantics, state estimation under delayed telemetry, and practical ML infrastructure.
- 📖 Designing Data-Intensive Applications (DDIA)/Boar book — learning how reliable systems handle data, scale, and failure reliably.
- 🧪 Evals for AI engineering — exploring how to measure, debug, and improve AI systems beyond "it seems to work."
I like to understand not just how things work, but why they break. ☕
Thanks for stopping by. Always something new on the workbench quenching the thirst to build.
Small improvements, carefully shipped.