I've been building Bespken — it joins your scheduled Zoom/Google Meet calls, transcribes them, and drafts a ready-to-send proposal or invoice from what was actually discussed, instead of leaving you to reconstruct scope and pricing from memory after the call ends.
The gap,
Otter, Fireflies, Fathom — they all give you a transcript and a summary. None of them give you the document you actually need to send to a client. That's the specific thing I'm trying to build toward.
How the pipeline actually works right now
- Connect Google Calendar (Supabase Auth + Google OAuth, calendar.readonly scope)
- See upcoming calls with video links on a dashboard
- Click "Join with Bespken" → a bot (via Meeting API) joins the call as a visible participant
- Call ends → webhook fires → transcript fetched from a presigned URL → sent to Gemini for structured extraction (client, scope, price, timeline, notes)
- Review screen shows extracted fields as editable inputs — with a mandatory, prominent warning to verify everything before use
- Generate a proposal document from the (possibly edited) data
This is fully working end-to-end in production right now, not just a mockup.
What I found testing it (the part that actually matters)
Before building the full thing, I ran two real test calls specifically to check the riskiest assumption: would transcription and AI extraction actually be reliable enough to trust for pricing data?
Marathi test call: failed badly. Confidence scores mostly under 0.5, numbers specifically got mangled.
English test call: numbers came through textually correct, but confidence scores were still surprisingly low (0.01–0.4) even on correct text. That's a real problem — confidence scores alone can't tell you which numbers to double-check.
Because of this, I made a deliberate call: no confidence-based auto-flagging yet. Instead, the review step is mandatory and the warning is loud, not a small footnote. I'd rather ship something honest about its limits than something that looks more trustworthy than it actually is.
What I'm still trying to figure out
This came directly out of a previous project (Screenshot Vault) where I built for weeks before discovering the core feature already existed for free in Google Photos. Learned that lesson the hard way, trying to apply it here — though I'll be honest, once the technical build got moving, I leaned more on my own conviction as a freelancer myself than I probably should have, instead of doing more outreach first.
So, genuinely: if you're a freelancer or consultant who quotes pricing on calls, I'd rather hear from you than just get a signup. What do you actually do right now after a call where pricing came up — and would something like this actually fit into that, or does it feel like solving a non-problem?
Landing page: https://bespken.vercel.app/ — but the honest answer in the comments is worth more to me right now.
Top comments (0)