Forge — Distributed Task Queue
In ProgressA fault-tolerant distributed job queue with a Raft-replicated coordinator and a chaos-test harness proving zero job loss across node failures.
GoRaftgRPCBadgerDBPrometheusDocker
- Raft-replicated coordinator with leader election and log replication for failover with no lost jobs.
- Effectively-once delivery via idempotency keys, retries with exponential backoff, and dead-letter queues.
- Jepsen-lite chaos harness that kills nodes mid-run and verifies zero job loss and no double-completion.
nnU-Net FCD-II Segmentation (Research)
Automated segmentation of FCD Type II lesions in 3D FLAIR MRI using nnU-Net.
PythonPyTorchnnU-NetNumPy
- Trained nnU-Net with 5-fold cross-validation on MRI from 85 FCD-II subjects (Dice 0.52 ± 0.05).
- Proposed automated axial FLAIR slice selection by peak voxel intensity, improving detection 30%.
- Published in Frontiers in Artificial Intelligence (2025).
A live SaaS that turns course syllabi into a calendar — built on a data-extraction pipeline.
Next.jsTypeScriptFastAPIRedisLLMs
- Shipped a live product (50+ users, 35+ extension installs) that auto-extracts deadlines from syllabi.
- Built an LLM extraction pipeline (95% accuracy) exporting to Google & Apple Calendar.
- Cut load times 60% with Redis caching.
Full-stack AI platform using OCR to convert 5+ note formats into structured markdown.
FastAPIPostgreSQLNext.jsTypeScript
- Built an OCR pipeline converting 5+ note formats to markdown at 90% accuracy.
- Created AI study tools (quizzes, flashcards, summaries) reducing prep time by 40%.
Centralized platform for student support services discovery
ReactFastAPIPostgreSQLJWT Auth
- Built platform for 80+ university services, attracting 1000+ visits & 50+ recurring users.
- Reduced student search time for support resources by 25%.
York University Calendar Project
Diversity calendar showcasing religious and cultural events
FastAPIReactMongoDBSeleniumLLMs
- Deployed diversity events calendar, adopted by 200+ faculty/staff.
- Automated content generation via scraping & LLMs, reducing curation effort by 90%.