The Problem
First-touch qualification calls were repetitive and ate hours of sales time each week.
What I Built
A Python agent that places real outbound calls over SIP/VoIP, transcribes the other person in real time with faster-whisper, decides what to say with an LLM (Groq while testing, a local Ollama model in production) and answers in a natural voice with Piper or voice-cloned F5-TTS.
Scheduling: A scheduler only dials within calling hours, so no calls go out at unsociable times.
Contact management: Google Sheets holds the contacts and the call log. Every outcome is written back for the team to follow up.
Voice cloning: F5-TTS can clone a specific voice from a short sample, so the agent sounds like a real team member.
The Outcome
Routine opening calls can run unattended, with every outcome written back for the team to follow up. Frees up hours of sales time each week for actual conversations.
0 likes · 0 views · Nov 15, 2026
Python agent that places real outbound sales calls over SIP/VoIP, transcribes in real time and responds in a natural voice.
Timeline
Jan 1, 2026 - Ongoing