R&D and operations of large-model training platforms — keeping training jobs fast, stable, and observable:

• Training platform — distributed training job scheduling, cluster resource management, multi-tenant quotas
• Observability — end-to-end job monitoring, alerting, and automated failure diagnosis
• Performance — GPU utilization, data-loading pipelines, checkpoint & resume

Toolbox

Volcano · Ray · Argo Workflows · Prometheus · Loki · PyTorch · Kubernetes · Docker

Research on intelligent signal processing for radar & communication signals:

• Signal detection & processing — Fourier analysis, time-frequency methods, communication-signal detection
• Modulation recognition — automatic modulation classification
• Vision-based signal intelligence — transforming RF analysis into computer-vision problems (spectrograms, constellation images) and solving them with deep learning

Flute player (8 years). Built Syrinx, an open-source flute practice app: score-following, sampled piano accompaniment, note-by-note pitch scoring.

TypeScript · React · Web Audio · ffmpeg