Their challenge: how do you let non-technical users explore complex data without building a new interface for every question?
Founded by Luka Kacil and Ryan Buckley, Shovels serves a wide range of customers including sales teams, contractors, material suppliers, site selection teams, and financial institutions that need answers from messy construction data but don't have engineering teams on standby.
Shovels sits on a goldmine of construction intelligence. They collect, clean, and unify fragmented building permit data from across the United States turning chaos into a comprehensive database that answers questions like:
Additionally, Shovels also offers four ways to access this data:
The first two served technical users well.
But Shovels Online, designed for non-technical users like sales teams, had a fundamental limitation: every use case needed its own custom interface.
"You always need to build the interface for a specific use case. You need to test it, interview users, and iterate. That's hard to do when people want answers that don't fit predefined dashboards."
~ Luka Kacil, Co-founder and CTO
For a small team managing multiple products, an API, enterprise workflows, and a web app this was unsustainable.
Ad hoc questions that didn't fit existing dashboards required engineering resources, customer interviews, and iteration cycles.
The real problem wasn't the data. It was making that data explorable for people who wanted quick answers and didn't have the technical expertise.
Shovels experimented with chat-based interfaces connected directly to their database through a local MCP (Model Context Protocol) setup.
The idea was simple: let users ask questions in natural language, and let the AI query the database directly.
Early results were promising, but the experience quickly became inefficient.
A sales team might spend an entire conversation refining a query, for eg. finding solar contractors in LA with specific permit values. They'd get their results, export them to their CRM, and move on.
Next month, they'd need similar data with slight tweaks.
"Do you go in and do the whole thing again? It becomes cumbersome. It's not as frictionless as a clean UI that's built exactly for this use case."
~ Luka Kacil, Co-founder and CTO
Construction data is inherently visual. Trends over time, geographic distributions, comparative analysis, these need charts and graphs for better visualization and understanding.
"People going in and analyzing, they need to see some charts to make sense of this. You want to see trends over states over the past decade. You can get this in table format. But it's not great, it's just not the best visualization."
~ Luka Kacil, Co-founder and CTO
Chat returned text and tables, real, actionable insights stayed buried.
Perhaps most frustrating: the AI would ask follow-up questions. Sometimes five or six in a row.
"You just feel overwhelmed. Now I have to answer with one long paragraph before we can continue. There's so much friction in this and the pace just stops."
~ Luka Kacil, Co-founder and CTO
While incorporated with the promise of speed and flexibility, chat resulted in more typing, waiting and as a result increased friction.
Shovels needed a way to combine the flexibility of conversation with the clarity and speed of a structured interface.
The turning point came when Luka discovered C1, our Generative UI API.
Instead of forcing users to type out every detail, generative UI could present structured choices: buttons, multi-select forms, dropdowns.
The AI still drove the conversation, but now it could show options instead of asking for them in text.
"When I saw the generative UI, especially when it came to the forms and all these interactive elements where the answer or selection was already provided to the user, that really clicked. This is a really smooth user experience."
~ Luka Kacil, Co-founder and CTO
"This is the middle ground. It's going to cater to all these ad hoc use cases that might not be so frequent but very important, where people go in, deep dive, and they get the data out that they need."
~ Luka Kacil, Co-founder and CTO
The technical implementation was straightforward.
The team followed Thesys's recommended path and integrated C1 largely out of the box. Their focus shifted from integration challenges to building the right workflows.
For Shovels, the biggest transformation was removing friction from how users interacted with the rich data they had.
The contrast with text-only chat is stark:
"What makes this a 10x experience is not just chat. It's the UI. Buttons, choices, visual feedback. You don't have to stop and write a long paragraph just to move forward."
~ Luka Kacil, Co-founder and CTO
By replacing text-heavy follow-ups with structured interactions, Charlie keeps users in flow and makes complex data feel approachable.
With C1, Charlie now enables:
Shovels is continuing to expand Charlie as a dynamic analytics surface for exploration.
To prepare Charlie for broader adoption, the team is adding essential supporting features:
Users can return to previous conversations and pick up where they left off, solving the "start over" problem from earlier chat experiments.
When users request large datasets, Charlie can:
The team is identifying the "minimum viable product" feature set needed to make Charlie worth charging for, making it a core product offering.
These workflows are designed to support performance, scalability, and cost control as the product moves toward broader adoption.
For Shovels, Generative UI has become the bridge between powerful construction data and a way their customers can actually use it.
"Chat is nice, but generative UI is the upgrade. It's what turns a powerful backend into something people want to keep using."
~ Luka Kacil, Co-founder and CTO
See how Generative UI can accelerate your AI product with Thesys
Thesys can help you save months of development, cut costs, and deliver interfaces your users trust.