1. Multilingual enquiry desk (schemes / policies / claims)
Citizens and customers ask about schemes, policies or claims in many languages; call centres can't cover them all.
Trigger: WhatsApp / Call
How the workflow runs
- Trigger
Enquiry in any language. A WhatsApp or Call trigger accepts the question as spoken/typed.
- Agent
Understand + respond in-language. An Agent node works in the caller's language (Sarvam covers 22+ Indian languages).
- Knowledge
Ground in official docs. A Knowledge node answers strictly from the scheme/policy documents — with citations, no guessing.
- Approval
Escalate sensitive cases. An Approval node routes claims/KYC/eligibility decisions to a human officer.
- Output
Confirm + record. An Output node summarizes and logs the interaction.
Channels & connectors
- Voice / Call
- Sarvam (22+ languages)
- Knowledge base
- Approval / HITL
Outcome
Anyone gets accurate, in-language answers; officers handle only the decisions that need judgement.
Why it helps
Extends coverage to every language and hour, while keeping every consequential decision with a human.
Build spec
1 agent — 1 in-language service agent; the sensitive-case escalation is an Approval node.
System prompt (paste-ready)
You are a citizen/customer service agent for {{organization}}. Work in the language the person uses (Sarvam covers 22+ Indian languages). Answer questions about schemes, policies and claims strictly from the official documents in Knowledge, with citations — never state an eligibility rule, entitlement amount, or policy exception you can't cite. Route any claim decision, KYC, or eligibility determination to a human officer. Be clear, patient and neutral.MCP connectors
- WhatsApp / Voice — conversation channel
- Sarvam — Indian-language voice/translation
- Case/record system — via a Tool (MCP) node
Built-in tools
- knowledge_search (official scheme/policy/claims documents)
Guardrails
- Grounded-only with citations — no invented eligibility rules, amounts or exceptions
- Claims, KYC and eligibility decisions always escalate to a human officer via the Approval node
- Every interaction is summarized and logged for audit
Output & delivery
The agent answers in the citizen's language from official docs with citations; sensitive decisions route to a human officer via the Approval node, and an Output node summarizes and records the interaction.
