projects
Research software for science, and systems I've built along the way.
research
AgileLens
Scalable CNN pipeline for strong gravitational lens discovery in Euclid imaging — data sourcing, cleaning, iterative VGG16 fine-tuning, candidate ranking and expert inspection. Found 130 new high-confidence lens candidates.
Astronomy research platform
Full-stack platform combining pretrained CNNs, galaxy datasets, RAG and data-processing tools over 200 GB+ of data, with CPU parallelism and CUDA acceleration; sped up researchers' object-detection workflow by 30%.
Image defect detection
Sliding-window algorithm that flags corrupted Euclid images, improving rejection accuracy by ~20% over the default Euclid catalog filtering.
Human-in-the-loop labeling tool
Annotation and visualization tool for galaxy image labeling and large-scale candidate validation, enabling fast iteration between model predictions and expert review.
Galaxy-object database
Searchable PostgreSQL + React database supporting CRUD and discovery workflows; cut manual data-retrieval time by 60%.
Research-docs RAG assistant
Natural-language retrieval over internal research documentation, plus reusable data-processing / training / plotting modules deployed for 20+ researchers at NASA, Caltech/JPL and the U.S. Euclid team.
engineering
Philo Homes backend
TypeScript/Node.js (Medusa v2) e-commerce backend for an AI room-design app; GCP → AWS migration cut cloud cost 40%, catalog sync went from 2 hours to 3 minutes, and OIDC-based CI/CD cut deploys from 30 to 5 minutes.
Distributed online shopping system
Spring Boot / MyBatis / MySQL service with AWS SQS for burst traffic, Redis + Lua distributed locking and the TCC pattern to prevent overselling during flash sales; CI/CD on ECS with CDK and CodePipeline.