$whoami

andreas demosthenous

cs undergrad @ university of cyprus
software engineer intern @ kios research and innovation center of excellence

about

i build things across the stack: geospatial python tooling at work, embedded firmware on weekends, the kernel of my own phone in between. comfortable somewhere between a syscall trace and a docker-compose file. interested in what large language models cost in joules once you're the one paying the power bill.

currently

›building python / qgis tooling for geospatial water-network modelling at kios
›writing an undergraduate thesis on the energy cost of llm serving stacks
›running baseplate/, my self-hosted docker stack, on a raspberry pi 5

interests

llm systems & on-device inference linux internals self-hosted infra embedded & iot geospatial python

projects // pinned

baseplate/ oss · single-host self-host

opinionated docker-compose stack for a single host: traefik with letsencrypt via cloudflare dns-01, cloudflare tunnel for outbound-only ingress, and opt-in profiles for nextcloud, vaultwarden, paperless-ngx, mosquitto, wireguard. the recipe i run on my pi 5, packaged so you can clone it and have your own homelab up before lunch.

dockertraefikcloudflare tunnelsself-hostedraspberry pi

smart_home_iot/ private · multi-node

multi-node home automation and a custom-built security system across esp32 and raspberry pi units; full-stack web ui for monitoring and control.

esp32raspberry pimqttfull-stack web

android_kernels/ downstream

maintained device trees and kernel forks for xiaomi (lavender, sdm660-common) and samsung (universal7880) devices on aosp / lineageos. low-level dive into arm bring-up, vendor blobs, and kernel patches.

aosplineageoslinux kernelarm

ollama-flake/ nix · oss

a nix flake packaging ollama for reproducible, isolated deployment of local llm inference. used to keep a llama.cpp / ollama stack pinned across machines.

nixollamallama.cpplocal llms

andreas demosthenous

about

currently

interests

projects // pinned

contact