AI-Powered Crop Advisories with Elastic Agent Builder

Join the Elastic Dallas User Group on Tuesday, June 16th for an exciting meetup.

We’ll feature a presentation from Ankit Hemant Lade (Data Scientist)and Sai Krishna Jasti (AI Engineer) followed by networking, refreshments, and pizza with the Dallas tech and Elastic community.

📅 Date: Tuesday, June 16th from 5:30-7:30 PM CT
📍 Location: Improving Dallas - 5445 Legacy Dr #100, Plano, TX 75024 (Google Maps)

🎤 Want to present at our next meetup? Email us at meetups@elastic.co
Are you interested in presenting your Elastic use case? We welcome 5-10 minute lightning talks, 45-minute deep dives, and everything in between. If you're interested, please send us an email at meetups@elastic.co.

📝Agenda:

• 5:30 pm: Doors open; say hi and eat some food.
• 6:00 pm: Building FarmSense: Hyper-Local Crop Advisories with ES|QL, ELSER, and Elastic Agent Builder - Ankit Hemant Lade — Data Scientist and Sai Krishna Jasti — AI Engineer
• 6:45 pm: Networking & refreshments
• 7:30 pm: Event ends

💭 Talk Details:
Building FarmSense: Hyper-Local Crop Advisories with ES|QL, ELSER, and Elastic Agent Builder - Ankit Hemant Lade — Data Scientist and Sai Krishna Jasti — AI Engineer

FarmSense is an AI agronomist that delivers hyper-contextual crop advisories to smallholder farmers in plain language — via Kibana Agent Chat or a Telegram bot. In this talk, we'll walk through how we used Elastic Agent Builder to orchestrate an LLM across seven custom tools that reason in parallel over four independent data streams: geo-aware climate time-series, regional pest outbreaks, ISRIC-style soil profiles, and FAO/CGIAR agronomic guidance. We'll dig into the architectural trade-offs — why we leaned on ES|QL for geo and time-series aggregations, how ELSER v2 bridges the "leaves look weird" → "chlorosis" semantic gap, and how Elastic Workflows handle audit logging and critical pest-risk webhook alerts. Expect a live demo, a tour of the 4-step pipeline, and a few lessons learned from shipping an agent that needs to respond end-to-end in under 60 seconds.