petercheng.dev · LinkedIn · Email · Résumé
I build backend systems and ML pipelines at the intersection of reliable engineering and applied research: distributed request pipelines and LLM infrastructure at Undergraduation.com, sensor-based depression detection at Syracuse, and medical-imaging segmentation at WashU’s AI for Health Institute.
- Distributed SSE + serverless request pipeline — 50+ concurrent users, zero blocking
- Fault-tolerant LLM gateway with failover & latency-based A/B routing — 99.9% uptime
- PostgreSQL composite-key caching — −40% redundant queries
- Row-Level Security + middleware — zero cross-tenant leakage on vector search
- Meta Sapien + OpenCV pipeline — segmentation masks across 10,000+ clinical frames
- FFmpeg high-speed extraction — −60% preprocessing latency on 50+ hrs of footage
- Dynamic exposure workflow — +80% human detection in low-light scenes
- Android multi-modal IMU collection framework (Kotlin, Factory pattern)
- Dockerized ML training — environment setup 2 hrs → 5 min
- Hybrid CNN-LSTM on high-frequency IMU — 87% depression-biomarker accuracy
- HIPAA-aligned data handling with clinical partners
- XGBoost sales forecasting with lag features — 92% prediction accuracy
- Automated ETL against the data warehouse — −40% extraction latency
- High-dimensional transaction analysis for Shanghai-region strategy
| # | Project | Stack |
|---|---|---|
01 |
Patient-Monitoring Vision Pipeline — overhead clinical video, sequence-error detection | Python · PyTorch · Meta Sapien · OpenCV · FFmpeg |
02 |
Engineering Course Scheduler — prereq-aware, conflict-free timetables | React · Node.js · AWS |
03 |
ATS — Application Tracker — Gmail ingestion + AI triage · 2nd place, Syracuse Tech Challenge | React · Node · PostgreSQL · Gmail API |
04 |
IMU Depression-Signal Collector — scheduled multi-modal sensing for Salekin Lab | Android · Kotlin · Signal Processing |
05 |
Library VR — gesture-driven campus library onboarding | Unity · C# · VR |
petercheng.dev · GitHub · LinkedIn · Email · Résumé
