Sofía
A WhatsApp sales agent that hears, sees, and remembers the conversation.
The problem
An insurance agency in Latin America needed to answer pricing questions, check availability, and book advisory calls over WhatsApp — around the clock, without losing the human touch or the thread of a conversation that arrives as three separate voice notes.
The solution
Sofía is a WhatsApp/Telegram agent that transcribes voice notes and reads images before responding, buffers rapid-fire messages so three quick texts become one coherent reply instead of three disjointed ones, checks pricing and FAQs from a live spreadsheet, books advisory calls directly on the calendar, and hands off to a human the moment a conversation needs one.
- 01Multimodal intake: voice notes are transcribed and images are analyzed before the message ever reaches the agent.
- 02Message buffering: rapid, back-to-back messages are held briefly and merged, so the agent replies to a full thought instead of a fragment.
- 03Conversation memory: a persistent store keeps context across the conversation, not just the last message.
- 04Live tools: pricing and FAQs are pulled from a spreadsheet in real time, and advisory calls are booked directly on the calendar.
- 05Human handoff: a dedicated escalation path hands the conversation to a person the moment the agent shouldn't handle it alone.
The result
Built and running against real CRM, calendar, and spreadsheet integrations — currently in gradual rollout as the team validates conversation quality before a full switch-on.
Validated end-to-end against real systems; rolling out gradually rather than an all-at-once launch, which is the standard approach for a client's primary sales channel.