speech-edge

Low-power speech-to-text edge router with local/cloud cascades, Kafka transport, OpenTelemetry metrics, and k3s deploys.

Personal long-term side project to explore real-time observability, power tradeoffs, and production-grade infra.

---

Overview

• Rust (edge-router) - low-latency GPIO handling async Kafka producer
• Go (stt-wrapper) - simple cgo bindings to whisper.cpp
• Python (power-sensor) - fast prototyping for INA219 power sensor & metrics
• Kafka + k3s - back-pressure support, self-contained cluster demos
• OpenTelemetry + Grafana - end-to-end traces + p95/p99 latency tracking

---

Prerequisites

Install core toolchains via Homebrew:

brew install go python@3.11 rustup-init kafkacat docker docker-compose

Then launch Docker:

open -a Docker
docker run --rm hello-world
docker compose version

---

Language Setup

Rust

Install via rustup:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
source ~/.cargo/env
rustup component add clippy

Quick sanity check:

cargo new hello && cd hello && cargo clippy

Python

Use pyenv for version control:

pyenv install 3.11.9
pyenv global 3.11.9
python --version

Go

No extra setup needed beyond brew install go:

go version

---

Code Hygiene with pre-commit

Create a .pre-commit-config.yaml file with the necessary hooks. Then run:

pre-commit install
pre-commit run --all-files

This installs and runs the following linters:

• Python - black, ruff, mypy
• Rust - clippy, rustfmt
• Go - go-fmt, go-vet, go-test-mod, go-mod-tidy

---

Other Notes

Kafka Deployment Notes

This project uses Kafka with ZooKeeper mode for CI and local development to ensure maximum compatibility and reliability across environments.

KRaft (Kafka Raft Mode)

• Kafka is transitioning towards a ZooKeeper-less architecture known as KRaft mode, which offers simplified deployment, faster start-up, and a native quorum-based controller management
• This project does not currently use KRaft by default. Future branches may include KRaft support.