Agrilink
About the
Project
Agrilink needed to scale stand and ticket sales for large trade fairs like Nación Berries, across databases of more than 1,500 contacts. They relied on decision-tree bots and a lot of manual work: the system broke the moment someone went off script, lost context when a lead moved from email to WhatsApp, and let high-value leads go cold.
We replaced the patches with a commercial operating system. We designed Sebastián, an autonomous sales agent on LangGraph and GPT-4o, wired into HubSpot to read and update the pipeline in real time, Microsoft Graph for corporate prospecting, and the WhatsApp API. Vector memory on Supabase let us retrain him in record time to pivot from selling B2B stands to B2C tickets without hallucinating.
- ClientAgrilink · Nación Berries
- Year2025
- RoleCognitive Architecture (LangGraph / GPT-4o) / Supabase RAG / HubSpot, Microsoft Graph & WhatsApp Integration
- Website
The Result
Agrilink went from a static-response bot to a proactive prospecting machine capable of holding hundreds of simultaneous negotiations. Sebastián qualifies leads, handles objections with long-term memory, and executes the handover at the exact moment of highest buying intent: booking the meeting in the human team's calendar and handing them a detailed client brief. The sales team stopped chasing conversations and now works only on pre-qualified deals.
