I am building a voice agent for the following use case:
Use case: [one-sentence description]
Call direction: [Outbound / Inbound]
Primary language: [e.g., Hindi + English code-mixed]
Agent persona: [Name, role, organisation – e.g., "Priya, customer support agent at Acme Bank"]
Agent gender: [Female / Male – this affects Hindi verb endings]
Tone: [e.g., warm, professional, empathetic]
The agent must handle these call types / flows:
1. [Flow 1 — e.g., "Customer wants to pay the EMI"]
2. [Flow 2 — e.g., "Customer disputes the amount"]
3. [Flow 3 — e.g., "Customer asks for a callback"]
Top 5 edge cases to handle:
1. [e.g., Wrong person picked up]
2. [e.g., User is busy]
3. [e.g., User becomes hostile]
4. [e.g., User speaks a different language than expected]
5. [e.g., User asks if this is an AI]
Generate a voice agent global prompt with the following sections:
- Role, Tone, Objective (one line each)
- Interaction Guidelines (response length, one-question-at-a-time rule,
anti-repetition, identity disclosure, hostility handling, ambiguity handling,
system confidentiality)
- Language Style (address forms, code-mixed words to preserve, spoken language norms)
Keep instructions in English. Write behavioral intent — do not script exact sentences
unless they are legally required. Aim for instructional style, not deterministic.
