I take AI workloads from a GPU in a rack to an endpoint your product can call.
Acting CTO at Peregrine Ventures AI • GPU Kubernetes • vLLM serving • compliance platforms that pass audits
- AMLVault — AML/CTF compliance platform for regulated firms. ~500 users across ~200 firms, in production since March 2026. amlvault.au
- Peregrine GPU compute — bare-metal GPU hosts, Proxmox passthrough, and vLLM serving for rental and internal workloads.
- Deplexify — agentic code-to-cloud deployment. Final-year project, paper under review.
Fixes merged into projects a lot of people depend on:
| Project | Stars | What I fixed | PR |
|---|---|---|---|
| postcss | 29k | Escape </style> in non-Root node output |
#2163 |
| coredns | 14k | Stopped holding the zone lock across a network lookup, a production hang open 17 months | #8601 |
| hyperium/http | 1.4k | Rejected a header-smuggling character in the HTTP/2 table | #876 |
Open and in review in tokio (33k), aiohttp (16k), and serde_json (5k). Reported an XSS in minify, now fixed upstream.
Every PR I've merged into other people's projects →



