Ship Lab
Making a complicated spaceship build easier to reason about.
TOOLS / A PERSONAL PROJECT
Comeet
From a recorded conversation to something you can work with.
In development · Desktop application · case studyWhere it started.
A transcript is useful, but returning to a long conversation still takes work. Comeet explores the next step: a desktop workspace for finding what was discussed, who spoke, and which actions were mentioned.
What I’m exploring.
The application connects audio import, speaker-labelled transcription, topic timelines, summaries and action items. Electron and React provide the desktop interface; transcription and analysis use external services, with Postgres for persistence.
A workflow, not just a generated summary.
The work spans importing recordings and existing transcripts, organizing conversations, revisiting topics over time, and extracting actions with their stated owners and deadlines. Speaker labels distinguish voices; they are not verified identities.
Making a desktop product hold together.
The implementation includes separate main and renderer processes, a preload bridge, persistent conversation storage, connection setup and an update flow. These supporting systems matter as much as the first successful transcription.
Treating generated output as a draft.
Transcription and generated analysis can be wrong. Summaries and action points should be checked against the original conversation. Recording consent, access to conversations and what is sent to external providers are product constraints, not afterthoughts.
Where it stands.
A working desktop implementation with a release history. This page presents the product and engineering work, not a download recommendation. Distribution is under review. Transcription and analysis require external services; it is not an offline-only or on-device AI product.