The Elasticsearch Agent Builder Hackathon
Here's what the community built

Earlier this year, Elastic hosted the Elasticsearch Agent Builder Hackathon, a challenge inviting developers to build meaningful, working AI agents using Elastic Agent Builder. The goal was to combine a reasoning model with Elastic Agent Builder tools like Elastic Workflows, Elasticsearch, or Elasticsearch Query Language (ES|QL) to automate messy internal workflows, connect disconnected systems, and build impactful, domain-specific agents. The response was remarkable. Developers from around the world submitted projects spanning healthcare, security, regulatory compliance, and beyond.
Each participant identified a genuine pain point in their field; thought carefully about how agents, search, and orchestration could work together to address it; and built something that solves a real problem. The results show what’s possible when large language model (LLM) reasoning is combined with agentic speed and structure, including pipelines that detect drug safety signals in under 60 seconds, adversarial testing systems that get smarter over time, and duplicate-detection tools that save healthcare workers days of manual work.
Below, you’ll find blog posts from the top three winning teams covering the problem they chose, how they designed their agent architecture, and the lessons they learned along the way. Whether you’re new to Agent Builder or already building production systems, these posts are worth reading.
The winners
PHAROS: Four agents, 60 seconds, one missed drug-safety signal away from disaster
Prajwal Sutar built PHAROS — a four-agent pharmacovigilance system that ingests FDA adverse event reports, runs WHO-standard statistical analysis entirely inside ES|QL, generates regulatory paperwork, and fires alerts to Slack, Jira, and email all in under a minute. His post covers the deliberate agent architecture, the decision to keep statistical computation inside Elasticsearch, and JSON-parsing in the pipeline.
Gauntlet: What happens when your agent’s tools fight back
Kavish Sathia built Gauntlet — an adversarial testing framework where a mocking agent intercepts your primary agent’s tool calls and tries to break it automatically and with a long-term memory that makes it more creative with every run. Rebuilt from scratch 48 hours before the deadline after a pivot, Gauntlet is a compelling argument for why happy-path testing isn’t enough for agents with real-world tool access. His post explains the two-memory architecture and how ES|QL’s completion function surprised him.
Catching invisible errors: A duplicate detection agent for Kenya’s HIV program
Fredrick Kioko is a solutions architect in Nairobi who builds health information systems across all 47 Kenyan counties. He brought the hackathon a problem he’d been watching compound for months: duplicate patient records in Kenya’s HIV testing infrastructure quietly inflating dashboards and wasting reagents. His three-agent system scanned 1,010 real anonymized records in under 10 seconds, surfacing 131 duplicates, including same-day multi-facility cases that would have taken weeks to catch manually. His blog illustrates why explainability isn’t a nice-to-have in clinical AI.
Learn more about the winners
These three projects represent very different domains, but they share a common theme: Each builder started with a concrete, costly problem and used Agent Builder to build something that reasons about it rather than just querying it. That’s exactly the kind of work this hackathon was designed to surface.
Read each winner’s blog below, and see what’s possible.
The release and timing of any features or functionality described in this post remain at Elastic's sole discretion. Any features or functionality not currently available may not be delivered on time or at all.
In this blog post, we may have used or referred to third party generative AI tools, which are owned and operated by their respective owners. Elastic does not have any control over the third party tools and we have no responsibility or liability for their content, operation or use, nor for any loss or damage that may arise from your use of such tools. Please exercise caution when using AI tools with personal, sensitive or confidential information. Any data you submit may be used for AI training or other purposes. There is no guarantee that information you provide will be kept secure or confidential. You should familiarize yourself with the privacy practices and terms of use of any generative AI tools prior to use.
Elastic, Elasticsearch, and associated marks are trademarks, logos or registered trademarks of elasticsearch B.V. in the United States and other countries. All other company and product names are trademarks, logos or registered trademarks of their respective owners.