( case · aurora · Tauá Resorts Group )
aurora: the AI concierge I built for the Tauá Resorts Group, from WhatsApp to Salesforce.
The official WhatsApp channel, an AI that consults the group's platform and bookings, and a handoff to the human queue in Salesforce, delivered end to end.
Results measured in production
- WhatsApp conversations
- 293k+
- resolved without human intervention
- 87%+
- average response time
- 2.1s
fact sheet
- client
- Tauá Resorts Group
- sector
- hospitality, 6 resorts
- role
- architecture, code and delivery, end to end
- period
- 2025 to Aug 2026
- stack
- TypeScript, Python, Next.js, Hono, Drizzle, WhatsApp Business API, Salesforce
What the numbers mean and what they don't
( limits )
87%+ is the rate of conversations resolved without human intervention. It does not measure guest satisfaction.
R$ 40MM+ is what the group's Black Friday grossed. The AI supported the e-commerce sales; it did not generate that revenue.
The numbers come from the Tauá Resorts Group operation, measured in production.
What I shipped for the Tauá Resorts Group
( what i shipped )
AI concierge on WhatsApp
I built aurora on the official WhatsApp Business API, with an AI that consults the platform and the events and bookings API and hands off to the human queue when it can't resolve. On the group's Black Friday, which grossed R$ 40MM+, the AI supported the e-commerce sales.
operations platform across 6 resorts
I shipped an operations platform replacing spreadsheets across the group's 6 resorts.
native AI inside Salesforce
I integrated native AI inside Tauá's Salesforce, without AgentForce, keeping control of the LLM stack, with custom apps for the operation.
aurora flow architecture
( architecture )
The path of a message, from the guest to the human queue, in six steps.
- 01
the guest writes
On WhatsApp, the channel they already use.
- 02
the official WhatsApp Business API delivers the message
The message arrives through the official API and moves on to the AI layer.
- 03
the AI understands and consults the platform
The AI consults the platform and the events and bookings API, with control of the LLM stack kept.
- 04
the AI plugs into Salesforce
Native AI integration, without AgentForce.
- 05
the AI answers in the channel
2.1s average response. 87%+ of conversations end here, without human intervention.
- 06
what the AI can't resolve goes to a person
Handoff to the human queue in Salesforce when the AI can't resolve.
Technical decisions
( decisions )
Salesforce without AgentForce
Keep control of the LLM stack.
own platform instead of spreadsheets
Activity schedules, spa, restaurants and QR codes for the resorts in one system that the AI consults.
own events and bookings API
The AI finds activities and books spa and restaurants directly, without going through an agent.
multiple integrations
Official WhatsApp, Salesforce (human handoff and marketing), bookings, checkout and payment, telephony and weather, all wired to the same concierge.
Timeline
( timeline )
2025
The project starts at the Tauá Resorts Group.
Aug 2025
aurora goes into production.
Feb 2026
I take on all of engineering, from architecture to delivery.
Aug 2026
End of the cycle.
What the client asked for
( context )
“Innovation needs to be translated into results, it needs to generate value for the business, drive sales.”
Fábio Estevez, CIO of the Tauá Resorts Group, to Exame (Mar 2025, translated)
( read at Exame )The client's ask was business results, and this case measures exactly that.