BeClear
BeClear, a video and audio product at Agency Enterprise: a Next.js + Tailwind + MongoDB backend with Mux uploads, AWS Transcribe and Google Speech transcription, SendGrid and Twilio.
- BPM
- 172
- GRADE
- B
- LEVEL
- 6
- RELEASES
- 0
- YEARS
- 2021–2022
Turning Speech Into Text, Twice
BeClear started three weeks into anotter and ran through the winter of 2021 until March 2022. The backend was Next.js 12 with Tailwind and MongoDB.
Mux handled uploads, so video and audio came in without me building an ingest pipeline. For transcription I wired up both AWS Transcribe and Google Speech. In 2021 the results really depended on the audio, the accent and the vocabulary, so having two meant we could compare them, and we weren't stuck if one changed its pricing or output. SendGrid and Twilio sent the notifications, because the whole flow was async: upload, wait, get told it's ready.
I was building this while anotter was ramping up, so I had a streaming pipeline and a transcription pipeline going at the same time. This is around when my title became Director of Technology Innovation, which mostly meant proving a technology worked and then handing it to the team that would run it.