Under The Hood
brainz isn’t some mystery box — it’s built like proper infrastructure. every layer is exposed, every call traceable, every piece swappable. run it in a container, rip it apart, rebuild it. your stack, your rules.
backend architecture
language: python 3.10+
core stack: fastapi + sqlalchemy + transformers + sentence-transformers
async-ready, tuned for llms. nothing blocking, nothing hidden.
brainz was designed for devs who hate waiting on “magic functions”. it’s clean, modular, and made to be torn open.
frontend stack
fast. reactive. no unnecessary bloat.
language: typescript
react – component-driven ui
vite – hot reload + fast builds
tailwind – minimal, utility-first styling
it feels native without trying to impress you with heavy animations. built for devs, not designers.
⚙devops & infra
full docker-native build. no weird global python installs, no backend/frontend version hell.
docker-compose up → you’re live.
works the same on your laptop or in prod. one command, no excuses.
model compatibility
brainz doesn’t care what transformer you throw at it, as long as it’s huggingface-compatible.
tested and running:
falcon
gpt-j
mistral
llama
anything else via
AutoModelForCausalLM
switching? change MODEL_NAME in .env, hot reload, done.
flexible & extendable
it’s all layered:
swap out vector engines
replace tokenizer logic
extend agents
build adapters for custom models
rewrite memory scoring if you want
brainz doesn’t fight you — it’s built to be messed with.
Last updated