Developer tooling / platform

Dockbench (Dockerized Tool Orchestration)

Dockerized CLI tools (music/AI/speech) orchestrated through compose, with a REST job API and MCP surface.

1 / 2

Challenge

Heterogeneous tools (Python scripts, GPU models, scrapers) each look different to calling software.

My Role

Built uniform REST job API (queue/poll/download) with API-key auth and JWT exchange, web UI, MCP server, and first-party CLIs (karaoke-cli, agent-cli, workflow-cli); backend for Singlio's audio pipeline (chord detection, source separation, transcription).

Architecture

Compose profiles per tool family; shared workspace mount; server/worker/Postgres/Redis stack calling tools via compose run.

Interesting decisions

Every tool looks the same over HTTP; caches under workspace/.cache to separate them from real output.

Result

Single API for 20+ tools across music analysis/synthesis, speech, image, and LLM workflows.

What I Learned

Compose profiles let one repo span machines with different hardware without forking.

Ask AI About This · All projects