Aimone Advisory Book a call
← All work
Delivered · mortgage broker

Every call summarized, and nothing promised gets forgotten

A lighter build than a custom app: every recorded call becomes a summary and dated follow-ups in a tracker the broker already used.

Book a free 30-minute fit call

What changed

The situation

A broker spends the day on the phone, and every call ends in small promises: send the pay stubs, check the rate, call back Thursday.

What I built

Calls record themselves

On the business line, with an announcement so everyone knows.

Private details are removed

Redaction happens before any AI reads the transcript.

Every call becomes a short summary

Plus the tasks it created: who owes what, by when.

It all lands in one tracker

A spreadsheet he can sort and share, kept in step with his loan system.

Why a pipeline and not an app

Not every problem needs custom software. This one needed the calls captured and the follow-ups written down, in a tool the broker already used. When the business outgrew the spreadsheet, the same ideas grew into Fundrail.

What was handed over

A working workflow, the architecture diagram, a written solution document, the tracker workbook, the extraction schema and the workflow file, so any competent person could run or change it.

How it fits together

Six steps in a row, run by one workflow tool, ending in a spreadsheet.

Call-summary pipelineTHE CALLPIPELINE · RUNS ON ITS OWNWHERE THE BROKER WORKS123456Business lineTwilio: calls recordedwith announcementn8n workflowruns each stepTranscriptionAssemblyAI, privatedetails redactedClaudefills a fixed schema:summary, tasks, detailsGoogle Sheetstracker: one rowper call and clientLoan systemkept in syncFollow-upsdated tasks,who owes whatBrokerreviews, calls back

Colors group the steps by kind: capture, workflow, AI and storage. Grey is outside the pipeline. Swipe the diagram sideways.

Step by step
  1. A call on the business line is recorded, and the workflow is told a new recording exists.
  2. The workflow sends the audio for transcription.
  3. Private details (account and ID numbers and the like) are redacted before the text goes anywhere else.
  4. Claude reads the redacted transcript and fills a fixed form (summary, tasks with owners and dates, key loan details), and the result lands in the tracker.
  5. The tracker stays in step with the loan system.
  6. Follow-up tasks sit where the broker works, ready for the callback.

Technical details

For your IT person: components, trade-offs and security. Open any section.

Components
LayerServiceWhat it does here
CaptureTwilioRecords calls on the business number with a spoken announcement, then notifies the workflow.
Orchestrationn8nRuns each step in order and keeps a log of every run.
SpeechAssemblyAITranscription with built-in redaction of personal information.
AIClaude APIStructured extraction against a written schema, so every call produces the same fields.
TrackerGoogle SheetsOne row per call, linked to the client, with open tasks and due dates. Synced with the loan system.
Design decisions and trade-offs
Redact before AIPrivate details are removed at transcription, so the AI step never sees them.
A schema, not a chatThe AI fills fixed fields, so every call produces the same kind of record and results are easy to compare.
Meet the user where he worksA spreadsheet he already trusted beat a new screen he would have to learn.
Everything handed overDiagram, solution document, schema and workflow file ship with the build.

Losing track of what was said on calls?

A pipeline like this is quoted at a fixed price. Book a free 30-minute fit call and I’ll tell you what it would take with your phone system.

Book a free 30-minute fit call

More write-ups

Client onboarding and portalsA private portal and tailored questionnaire for every new client, set up by one script.Read the write-up →This websiteMoved off a website builder onto my own AWS account.Read the write-up →How I buildThe pattern every client environment follows.Read the write-up →