OPERATING
TensorHost — running a GPU hosting fleet on Vast.ai with a small team.
- Building tensorhost-pricing (dynamic pricing)
- RentalDetector service shipped — store + runner layers
- Full unit-test coverage, more services queued
Running the fleet with colleagues. Building tensorhost-pricing in parallel — a dynamic pricing system, with the RentalDetector service already shipped (store + runner layers, full unit test coverage).
Running the Imperial College Robotics Society and the Electrical & Electronic Engineering Society in parallel. Plus Head of IPRL, the Imperial Planetary Rover Lab — ground rover and UAV subsystems.
RV64 on QEMU with OpenSBI. Phase 0 (bare-metal UART) complete; 8-phase roadmap through traps, paging, scheduling, FS. Written from scratch, in the open.
ANSI C frontend with Bison/Flex, written in C++. 42-test suite passing, CMake build. The course made me fall in love with compilers properly.
RKE2 with Grafana, Prometheus, PostgreSQL. Authentik SSO integrated across the K8s Dashboard (via oauth2-proxy) and ArgoCD (OIDC). Self-hosted Outline wiki on CloudNativePG + MinIO sits on top.
Headscale v0.28 + Headplane v0.6.2, Authentik OIDC, Nginx reverse proxy, Let's Encrypt TLS. ACL policy gates exit nodes and team access.
OPNsense router, Proxmox migration on HP ProDesk 400 G4 + EliteDesk 800 G2, GPU server BIOS flash, Pterodactyl game server. Plus a custom multi-server KVM — Raspberry Pi 3B + RP2040 Picos + relays + HDMI switch + PiKVM.
Algorithms, compilers, systems. Picked up a soldering iron in the same week. Haven't put either down since.
> If you're working on systems, compilers, or building real infrastructure — or you've got an excuse to put a GPU somewhere weird — I'd love to hear about it.
Currently based in West London. Reply within a day, usually.