Snowflake Select Tier Partner AI Hakathon

Accelerating Innovation: Diving into Snowflake Cortex AI

Last week, some of our consultants had the opportunity to join the Snowflake Partner AI Hackathon in New Zealand—a dynamic, hands-on event designed for pre-sales, data, AI, and consulting professionals for Snowflake Select Tier partners. It was a day of practical learning and creative collaboration, building an AI solution for a real-world use case.

Bringing Cortex AI to Life

The hackathon wasn’t just theory—it offered real-world enablement to build business use cases using Snowflake Cortex AI on the Snowflake AI Data Cloud. From ideation through to pitch, attendees were encouraged to think big, while gaining direct experience with these cutting-edge tools.

Four Key Takeaways

1. Cortex AI is Ridiculously Easy to Use: Despite the variety of options and approaches, Snowflake’s Cortex AI proved intuitive and accessible.
2. Data is Still King: Clean, curated data remains essential—especially when working with generative AI.
3. Serverless  ≠ Costless: While Cortex runs on serverless compute, processing costs still apply. Sometimes, patience is more valuable than performance.
4. AI is Rapidly Evolving: Not all the latest models are available in Asia-Pacific yet, and Snowflake is actively rolling out new tools.

The Hackathon Journey - A Day Packed with Momentum

Participants were split into groups and given Matariki-themed use cases to choose from. It was competitive and our group started by asking Google’s Gemini which scenario was easiest—Fishing came out on top.

We then imagined a scenario where an overseas tourist might want to visit New Zealand for a fishing holiday and asked the question: “What are the best days to catch fish?”

Using Snowflake’s resources, we ran a simple query: SELECTAI_COMPLETE('llama4', 'question? || data') FROM (example_sql)

Just like that, Snowflake’s AI returned our answer instantly. We expanded the dataset to include tourist related event data, enabling more complex queries. We also spent some time building a Streamlit front end app for our tourist.

Iteration and Innovation

Realising our manual approach excluded valuable data and underutilised Snowflake’s capabilities, we pivoted:

• Streamlit became our chatbot interface.
• It called a Snowflake API, which triggered a Snowflake Agent.‍
• The agent chose between Cortex Analyst (structured queries) and Cortex Search (unstructured files like fishing regulations).‍
• Cortex Analyst used a Semantic Layer to generate SQL.
• Cortex Guard filtered inappropriate responses.

This approach took a few more hours. We spent the final hour debugging Python—Cortex Analyst only generates SQL, it doesn’t execute it (unlike the manual interface).

Adding new data was simple: just add tables to the Semantic Layer using the Snowsight wizard or YAML.

Final Solution

We expanded our tourist use case and prepared our data to answer our final tourist question: “I'm going to be in Auckland from 4–9 August. What’s the best day to go fishing and one event I can attend while I’m there?”

The AI responded: “5 August evening – based on the tide for fishing, and the Beer and Pie Festival on the 6th.”

We also asked about fish catch limits and sizes (it told us how to measure a snapper), and for directions to the event from the hackathon venue. Each question took around 10 seconds to answer.

With a bit of setup, data preparation, Snowflake AI is a power tool.

Lessons Learned

• Data quality is paramount: A CSV import error placed “City” data into the “Region” column, causing query mismatches. Fixable at the Semantic Layer.
• Patience is key: Teams ran out of Snowflake credits trying to speed up processes. Most AI time is spent in serverless compute, therefore the warehouse size doesn’t impact performance.‍
• Unstructured data is easy: We built an app called “Could I Eat That?”—based on a photo. Our app returned nutritional advice and recipes. One example was a poisonous mushroom!‍
• AI model considerations: Not all AI models are available in Sydney (at the time of writing). Some are only accessible cross-region (at higher cost), and models like ChatGPT-4 are only available in the US during early adoption phase with some consideration to data movement out of region until the models become available where your data resides.

Looking Ahead

Snowflake is releasing a pre-built app called SnowflakeIntelligence, which allows users to query data directly. It replaces the Streamlit/API/Agent components of our hackathon solution and uses the same AI Semantic Layer—making it a smart investment for future development.

As the AI and data space continues to expand, having deep experience with platforms like Snowflake Cortex AI will be invaluable—our investment is for our consultants' professional growth, as well as for driving impact across our clients' organisations.

Here’s to more hackathons, more innovation, and more futures shaped by intelligent tech! Talk to us today about how to implement and get the most out of Snowflake AI.

‍