AI voice agent for phone support and collections
An agent that talks on the phone in real time, negotiates within defined rules and produces an automatic audit of every call.
- Industry
- Telephony & conversational AI
- Status
- In development
- My role
- Solution architecture, data modeling and full functional specification
Collection teams spend hours on repetitive calls and have little visibility into the quality of each conversation. The goal is to automate first contact without losing control over what the agent can say.
I designed an architecture that connects a SIP trunk to a multimodal AI model with real-time bidirectional audio. The agent looks up the debtor, simulates offers within configured limits and hands off to a human when needed. After the call, an analytics copilot transcribes, separates speakers and assigns a 0–100 score with evidence and improvement tips.
- Real-time conversation with a natural voice through multimodal AI
- Per-agent guardrails and negotiation limits
- Human handoff and block-based visual IVR
- Post-call audit with score, evidence and coaching
Anonymized case: client names, domains and sensitive data have been omitted.