Choose your own data adventure: Winners from our global hackathon

Beyond dashboards: Meet the winners of our global Hex hackathon

What happens when you give data people a blank canvas, a few weeks, and the full power of Hex?

You get projects that feel less like dashboards — and more like living, breathing data experiences.

Earlier this season, we partnered with Devpost to host a virtual hackathon challenging builders to use Hex and publicly available datasets to uncover insights, test bold ideas, and build something that simply wouldn’t exist in a traditional BI tool.

Participants explored everything from podcast virality to zero-day risk, from public library funding to civic transparency. They stitched together APIs, wrangled messy datasets, and used AI to turn raw information into something you can actually use.

🥇 What's Lenny's Podcast Success Secret?

The idea: What makes a podcast episode go viral?

When Lenny Rachitsky made the full transcripts of their podcast public, this team saw an opportunity: analyze 285 episodes and reverse-engineer what separates the 10× hits from the rest.

Not a small task. Especially when your raw material is hundreds of long-form transcripts.

How they built it

• Parsed 285 episode transcripts with an LLM to extract structured details about guests, topics, products mentioned, and intro characteristics
• Pulled YouTube metadata and engagement metrics via API
• Mapped transcripts to performance data to identify patterns
• Used Hex’s notebook environment to wrangle, analyze, and shape everything into a fully interactive app

This wasn’t just analysis — it was ingestion, enrichment, modeling, and storytelling in one workspace.

What they found

• The top 5–10% of episodes earn 10–20x the median views
• Guests from high-profile companies and CEO/founder titles strongly correlate with breakout performance
• Product mentions evolve with the times — ChatGPT surged post-2023, newer tools are climbing fast, while some favorites remain evergreen
• Tuesday through Saturday releases outperform Monday and Sunday drops

And the best kind of insights: the non-insights.

• Episode length has zero correlation with views
• Subscriber growth doesn’t predict episode-level virality

The result feels less like reading a report and more like stepping inside the research — filter, compare, and test your own theories about what makes content resonate.

🥈 0-day Radar

The idea: Security teams are drowning in alerts. Everything is “critical.” But not everything is equally urgent.

This project asked a sharper question: what if we prioritized vulnerabilities based on the likelihood they’ll be exploited — not just how bad they look on paper?

How they built it

The builder created a “Risk Triangle” architecture that merges three sources:

• NVD (NIST) for base CVSS severity
• CISA KEV for confirmed exploitation in the wild
• EPSS for machine-learning-based probability of exploitation

Inside Hex, they used Python and pandas to compute a custom Composite Risk Score, heavily weighting real-world exploitation probability and confirmed activity over theoretical severity.

They also added an AI-powered “CISO Translator,” integrating an LLM via API to convert technical vulnerability descriptions into executive-ready summaries — with a rule-based fallback engine for reliability if the model failed to return structured output.

What they found

By visualizing vulnerabilities in a Risk Matrix (age vs. EPSS), the project surfaced a truth many teams feel but rarely see clearly:

• Many “critical” vulnerabilities sit untouched for years
• Meanwhile, newer, lower-severity bugs can be weaponized almost immediately

The output isn’t just a dashboard — it’s a prioritization system: a ranked “Kill List” that reflects real-world urgency, not just scary labels.

🥉 Public Libraries Peer-to-Peer Performance Analysis

The idea: Public libraries are community lifelines — internet access, job resources, after-school programs, safe spaces. Yet many operate with flat or shrinking budgets, outdated collections, and skeleton staff.

Here’s the brutal loop: libraries need funding to improve, but securing grants requires data-driven narratives to prove the need — and building those narratives requires time and analytical capacity that many libraries simply don’t have.

This project set out to break that cycle by democratizing benchmarking and storytelling for library staff.

How they built it

• Built a scalable warehouse in Snowflake
• Modeled data with dbt + semantic modeling in Hex
• Used SQL + Python in Hex for transformation and dynamic metric generation
• Integrated Groq API (llama-3.1) to generate data-driven narratives grounded in relevant metrics and context
• Used Hex notebooks, agent, and Threads to make the experience accessible to non-technical users

The end result is a dashboard that doesn’t just display data side-by-side — it does the analytical heavy lifting.

What they found

The “finding” here is less about a single headline insight and more about what becomes possible when analysis is made usable:

• Libraries can identify their true peers using customizable criteria and similarity scoring
• They can benchmark performance with per capita metrics that update dynamically as the peer group changes
• They can generate plain-language narratives to support grants and advocacy — just by asking questions

It’s a practical, scalable tool designed around real constraints — and it meets library staff where they are.

🎗️ Civic Impact Compass

The idea

Congress introduces 10,000+ bills per session. Most die in committee. A handful become law.

But the bigger problem is simpler: most people don’t know which bills affect their healthcare, taxes, education, or daily lives — even though the data is public.

Civic Impact Compass flips the usual legislative tracker on its head: instead of asking you to search for bills, it starts with you (your state, your interests) and surfaces what’s relevant — with plain-language explanations.

How they built it

• Pulled live data from the Congress.gov API
• Built an analysis pipeline in Python + pandas inside Hex to normalize nested JSON and categorize bills into citizen-relevant topics
• Added a personalization engine via input parameter cells (interest area, state, bill status, activity windows)
• Used NetworkX to build bipartisan co-sponsorship graphs and calculate betweenness centrality to identify “bridge builders”
• Visualized networks with PyVis and activity patterns with Plotly
• Used the Hex agent to generate bill deep-dive cards and plain-language explanations tailored to the filtered dataset
• Enabled Threads so users can ask natural-language questions (“Which bills should small business owners watch?”) and get answers grounded in the actual data

What they found

This app surfaces patterns that are hard to see in traditional trackers:

• Bipartisan collaboration is measurable — and “bridge builders” emerge clearly when you analyze co-sponsorship networks
• Legislative activity isn’t just volume; advancement matters. Their productivity scoring rewards bills that actually move (and especially become law)
• The most valuable output is personalization: instead of 10,000 bills, you get the 10–50 that actually matter to you, explained in plain language

It’s civic transparency that feels usable — not overwhelming.

Why we run hackathons like this

The theme of this challenge was simple: build something that wouldn’t exist in a static dashboard.

Across submissions, we saw notebooks turn into shareable apps, APIs stitched together with semantic modeling, and LLMs used as analysts and translators — all in one collaborative workspace.

The best projects weren’t just technically impressive. They were curious, opinionated, and human.

That’s the future of analytics we believe in — where analysis, AI, and application-building live in the same place, and where anyone with a question can turn it into something interactive and shareable.

To everyone who participated: thanks for building with us.

Want in next time? Keep an eye on hex.tech/events.