星尘纵易散，银汉自长盈。清辉苍穹倾，同歌寥廓永夜明。
Undergrad (class of 2028) passionate about Cloud Native infrastructure and High Performance Computing. Currently working at 39 AI / FishAudio as an Infrastructure Engineer, specializing in GPU cluster operations at scale.
Previously an Audio Algorithm Engineer focusing on TTS/SVC model development. Transitioned to Infrastructure to build the systems that power AI — from bare-metal GPU clusters to Kubernetes-orchestrated training pipelines.
- Compute — Operate 100–1,000+ GPU training clusters on bare metal and cloud, scheduled with Slurm (training) and Kubernetes (services & MLOps).
- Storage — Run petabyte-scale Ceph (object storage + CephFS) that feeds our training data pipelines.
- Network — Run our own ASN AS402567 with a BGP edge and 100Gbps+ public uplinks.
- Automation — Bare-metal provisioning through day-2 operations with Ansible; observability with Prometheus and Grafana.
All of it is the substrate beneath FishAudio's production TTS models.
Speech synthesis projects I've built and contributed to
| Project | Description | My Role | Stars |
|---|---|---|---|
| Bert-VITS2 | VITS2 backbone with multilingual-bert | Top contributor |  |
| fish-speech | SOTA Open Source TTS | Contributor |  |
| so-vits-svc | SoftVC VITS Singing Voice Conversion | Contributor |  |
Also contributed to ZeroBot-Plugin and fish-diffusion.
"In the world of Infra, every star matters."
Building the infrastructure for the AI era