Josue Arango
All projects

2024 · Used in production · AI & Agents

AI Sales Call Follow-up Agent

Speech-to-text · LLM · automated client emails

Sales calls at a multi-client IT services company were saved as MP3 files and then forgotten. This pipeline transcribes every recording with a Whisper-class speech-to-text model, passes the transcript to an LLM that pulls out what the client asked for, the products discussed and the agreed next step, and sends the client a personalised follow-up email automatically, so no call ends without a written follow-up.

Technologies

PythonWhisper (speech-to-text)LLMEmail APIREST APIs

My role

Sole author. Built the recording ingestion, the transcription step, the LLM extraction prompt and the automated follow-up emails.

Highlights

  • Batch transcription of MP3 call recordings
  • LLM extraction of client needs, products and next steps into structured fields
  • Written follow-up for 100% of sales calls, sent the same day
  • Removes the manual note-taking and follow-up work from the sales team

Screenshots and diagrams

Call recording, transcription, LLM extraction, personalised follow-up email
Call recording, transcription, LLM extraction, personalised follow-up email