AI agent for calling companies
AI & Machine Learning
Demonstration stand of a voice AI agent that autonomously calls companies (in the demo — via the browser microphone) and conducts a conversation based on a cold call script: introduces itself, determines relevance and the decision-maker, requests contact information for a quote, and properly handles refusals.
After the conversation ends, its summary is automatically recorded in a results table: company name, status (interest / call back / refusal / wrong number), contact person, where to send the quote, call back time, a brief summary of the conversation, and its duration.
The main difference from a "model that just talks": facts are recorded by code, not the model. Email is noted only if it was actually mentioned by the interlocutor; call back time is noted only if it was provided. If the model decides "interest," but the interlocutor says "no need" — the interlocutor wins, and the discrepancy goes into the log. The agent's voice is synthesized by Piper locally on our server (Ukrainian and Russian), speech recognition occurs directly in the browser; there is a text mode without a microphone, which also works from a phone.
This is a demonstration stand, not a project delivered to the client and not real telephony: in the production version, a telephone provider (Twilio / Vapi / Retell) is connected instead of the microphone, while the agent's logic, script, and fact-checking remain the same. All companies, names, and phone numbers on the stand are fictional. #AI #voiceagent #calling #automation #chatbot
After the conversation ends, its summary is automatically recorded in a results table: company name, status (interest / call back / refusal / wrong number), contact person, where to send the quote, call back time, a brief summary of the conversation, and its duration.
The main difference from a "model that just talks": facts are recorded by code, not the model. Email is noted only if it was actually mentioned by the interlocutor; call back time is noted only if it was provided. If the model decides "interest," but the interlocutor says "no need" — the interlocutor wins, and the discrepancy goes into the log. The agent's voice is synthesized by Piper locally on our server (Ukrainian and Russian), speech recognition occurs directly in the browser; there is a text mode without a microphone, which also works from a phone.
This is a demonstration stand, not a project delivered to the client and not real telephony: in the production version, a telephone provider (Twilio / Vapi / Retell) is connected instead of the microphone, while the agent's logic, script, and fact-checking remain the same. All companies, names, and phone numbers on the stand are fictional. #AI #voiceagent #calling #automation #chatbot