Skip to content
~/patrick/projects/voice-to-crm-ios
← projects

system lane · 2026

Voice-to-CRM iOS App

A native iOS app that turns a consultant's spoken note after a visit into a CRM record, by transcribing it and extracting the company, contact and outcome.

pipeline
  1. data (covered)
  2. model (covered)
  3. eval (not covered)
  4. serve (covered)
  5. monitor (not covered)
result

Apple approved it and a TestFlight build existed, and it was cut from scope before it reached any consultants.

stack

SwiftSwiftUIAWS TranscribeAWS BedrockClaudeSAML SSO

I built this at TechForce Advisors for consultants who visit businesses door to door, so that they could log a conversation without typing it up afterwards. It passed Apple’s review and had a TestFlight build. The company cut it from scope in July, before it reached any consultants, so it never had users.

Architecture

iOS appAPITranscribeClaudeSave and sync records a noteNext.js, Amplifyspeech to textextracts fieldsDynamoDB, CRMsign-in is SAML single sign-on through the company portal

A consultant records a voice note in the SwiftUI app, which sends it to an API built in Next.js on Amplify. The API saves the audio to S3 and has AWS Transcribe turn it into text. Claude on Bedrock then extracts the company, the contact, the outcome and the products discussed, and writes a short summary of the visit. The API stores the visit in DynamoDB and syncs it to the CRM as a contact, a pinned note and an opportunity for each product discussed.

Sign-in

Consultants sign in with SAML single sign-on through the company portal. The app opens the portal’s login in a system web session, the portal posts a signed assertion back to the API, and the API hands the app a one-time code through a custom-scheme deep link. The app exchanges that code for a token and keeps it in the iOS Keychain.

Decisions and tradeoffs

  • The CRM is a downstream copy. DynamoDB is the record of each visit, and the CRM sync can fail without losing it.
  • AWS-native services. Transcribe and Bedrock keep the audio and the transcripts inside the company’s AWS account, instead of sending them to a separate AI vendor.
  • The model can only choose from known values. The extraction accepts only a fixed set of outcomes and products, so a bad answer comes back empty and not as free text that lands in the CRM.
nextSlack CRM Agent →

keyboard

⌘K /
Search and commands
t T
Next / previous theme
j k
Move through items on the page (k on the first one goes back to the top)
o
Open the selected item
g then h w p b a l
Home, Work, Projects, Blog, About, Lab
c
Copy email
?
This list

Choose a theme

Dark

Light

You can also press t to step through them.