Over 38 days from September 16 to October 24 in 2025, Google Cloud, Elastic, and Fivetran held a joint AI-themed hackathon. Each contestant was challenged to build a compelling, innovative technical product combining Google Cloud products, such as BigQuery, Vertex AI, and Gemini, with either Elastic’s hybrid search capabilities or Fivetran’s Connector SDK.
Each contestant was required to submit:
- The hosted project
- A URL to the open source code repository
- A 3-minute demo video
Overall, 2770 people participated, completing and submitting a total of 171 projects, of which 44 used Fivetran’s Connector SDK. A maximum of four individuals could form a team, though most consisted of one or two participants.
The top 3 winners for each of the Elastic and Fivetran challenges received cash prizes, with the first-place winners also receiving an opportunity for a social media promotion. The following projects made up the top 3.
First place: Alzora
The two-person team that built Alzora set out to improve the lives of patients experiencing cognitive decline and aid their caregivers.
Under the hood, Alzora is powered by a custom Fivetran connector that extracts data from a TiDB database and loads it into BigQuery, dbt transformations that turn the data into embeddings for Vertex AI, and Census rETL to move data back into operational systems.
With the power of this comprehensive architecture for data movement, Alzora includes multiple impactful features, including:
- Interpreting scans and data streams from wearables
- Automated alerts when the patient leaves a designated safe zone
- Locating patients through GPS coordinates
- Monitoring of vitals through wearables
- Acting as a memory assistant by allowing patients to save text and images
All of these capabilities are accessible through a natural language interface.
The connector for the TiDB Ddatabase built by the team has been contributed to our community connectors and is available for you to use here
Second place: DocuLytics
DocuLytics was inspired by the difficulty of transforming raw data from DocuSign envelopes into actionable insights.
The DocuLytics architecture consists of a custom Fivetran connector for DocuSign that automatically detects changes to envelopes and loads them into BigQuery. On the front end, DocuLytics offers interactive, customizable dashboards with metrics such as envelope status, contract cycle time, completion rates, and user behavior, as well as a chatbot that can run queries.
Ultimately, users of DocuLytics can expect conversational analytics using data about documents that teams in their organizations, especially legal and sales, have signed. They will be able to analyze contracts and related documents without manually assembling reports or dashboards. Legal teams can automatically detect compatibility issues between new contracts and existing compliance requirements, while sales teams can identify both high-value customers and sellers and automate reminders and similar tasks.
The DocuLytics team has contributed their DocuSign connector here.
Third place: MergeMind
Modern software development is often bottlenecked by code review, risk assessment of merge requests, and other project management tasks. Traditional tools show basic metrics but leave contributors and project managers guessing about important details.
The sole developer behind MergeMind built the project to automate many tasks related to merge requests. The project consists of a custom Fivetran connector for GitLabs and dbt transformations that turn the data into embeddings for vector search by Vertex AI. MergeMind then surfaces recommendations for the team in a dashboard.
Specifically, MergeMind assesses:
- Architecture
- Security
- Team throughput and risk profile
- Predicted merge time
- Potential issues
Instead of manually reviewing code changes, MergeMind summarizes what the code changes will do, then automatically evaluates them, identifies risk, and recommends improvements.
The future of AI is in the application layer
Each of this hackathon’s winners demonstrates how combining AI with thoughtfully curated datasets and well-designed architectures can create entirely new capabilities and categories of products. All the projects other than the winning entries were no slouches either, and included healthcare search and recommendation, preventive maintenance for airline fleets, energy use optimization, and more.
The results of this hackathon show that most of the practical value of AI won’t come directly from further advancement in foundation models, but from the application layer. By analogy, if foundation models like Gemini, ChatGPT, Cursor, and so forth are power plants, then the application layer is the vast array of machinery, appliances, and consumer goods powered by electricity. As the future unfolds, business innovation will hinge on the unique data sets that augment foundation models and the architectures built around them.
[CTA_MODULE]
