Engineering decisions, product tradeoffs, and what it actually takes to ship a real-time meeting AI. Pain and gain, written down as we go.
One per major feature. Pain that motivated the build, the engineering decisions made under fire, the outcome.
Latency budgets, ONNX choices, and the UX of never losing your place mid-sentence.
Date.now() vs meeting-relative ms. A one-line fix that unblocked an entire feature.
The math. Which tasks need Claude, which can run on local Llama, and how we pick in 50ms.
What it takes to keep conversations private when the product depends on understanding them.
Low volume. No marketing fluff. Just the engineering and product posts.
Or follow along on the beta waitlist at /download.