An AI company quietly turning America's most boring paperwork - the building permit - into the most useful dataset in real estate, climate, and construction.
It is Tuesday morning, and a permit clerk in Kern County, California is doing what permit clerks have done since the invention of the carbon copy - taking a stapled stack of architectural drawings and scanning them into a county portal that looks like it was built during the Clinton administration. The file lands in a folder. Nobody reads it.
Except, today, somebody does. Inside a quiet office park in Lafayette, an ingestion pipeline at a 39-person company called Shovels picks up the upload, normalizes the schema, extracts the parties involved, geocodes the address, classifies the work type, links it to a contractor profile that already exists in the database, and pushes the record - now structured, queryable, joinable - into Snowflake shares used by climate-tech firms in Brooklyn, building-materials suppliers in Atlanta, and a heat-pump installer in Sacramento who will, by Friday, have a list of every household within thirty miles that just pulled a permit for a new HVAC system.
That whole sequence took ninety seconds. The clerk has no idea.
Building permits are the first verifiable signal that anything in the physical world is about to change - a roof, a panel, a charger, a warehouse, a skyscraper. They are also, almost universally, a mess. Each of the roughly 22,000 jurisdictions in the United States runs its own portal, schema, naming convention, and PDF habit. Some publish CSVs. Some publish nothing. A few still ask you to drive in.
Shovels collects all of it. The company has indexed more than 130 million permits across 1,800+ jurisdictions, profiled 2.3 million contractors, and built coverage that touches roughly 85% of the US population. It pipes the result out through a web app, an API, a CLI, Snowflake, Databricks, BigQuery, and a natural-language interface named Charlie.
If that sounds like infrastructure work, that is the point. Shovels is not trying to be the consumer brand of construction data. It is trying to be the substrate underneath it.
Borrowed from the website's homepage. The phrase does the work of an entire pitch deck.
Shovels resists the temptation to make customers learn a new app. The data shows up wherever the customer already works.
Point-and-click exploration and CSV exports. Where the analysts start.
Programmatic access for CRMs, lead-routing, and custom apps.
Parquet drops and live shares to Snowflake, Databricks and BigQuery.
Ask, in English, what got permitted last month in your ZIP. Get an answer.
B2B and homeowner segments built from real permit signals.
QGIS, ArcGIS, and the rest of the spatial-analyst toolkit.
An agent-first command line for AI workflows. Designed for LLMs first, humans second - which is an interesting choice, and a deliberate one.
Python, dbt, DuckDB, Prefect, Postgres, FastAPI, Vue, Fargate, scikit-learn, and Claude. A modern data-engineering greatest-hits.
Every retrofit leaves a permit. Climate operators use Shovels to find the next install before the truck rolls.
Map demand at the ZIP-code level. Stop guessing which markets are heating up.
Permit activity is leading-indicator data for neighborhood movement and asset value.
Targeted leads based on actual filings, not direct-mail lists from 2014.
Site planning informed by what is being permitted around the parcel.
Companies like Beam plug Shovels in to enrich their own products.
Integration of Shovels data resulted in a remarkable 20 to 30% higher likelihood of contractor engagement. Beam, construction software customer
Shovels has a rare and massive opportunity to become the definitive source of truth for high-value, hard-to-access datasets at the local government level. Rexhi Dollaku, General Partner, Base10 Partners
A repeat operator who picked the unsexiest possible dataset and ran at it. Buckley is the one publishing under ryan@shovels.ai and writing the long blog posts about why local-government PDFs deserve more respect than they get.
Runs the ingestion pipeline that eats 5 million records a month and keeps the schema honest. The reason 1,800 jurisdictions look like one table.
Led by Base10 Partners. Used to expand the team, scale ingestion, and push into adjacent local-gov datasets.
Earlier participation from Coelius Capital and angels.
Pull a feed of every solar, heat-pump, or EV-charger permit in your service area, the day it's filed.
License status, permit history, jurisdictions worked, recent activity - all from one endpoint.
Build TAM models on actual permit volume instead of survey extrapolations.
Cross-reference filings against contractor licenses and historical patterns.
Layer permit density over parcels in QGIS or ArcGIS.
Charlie answers questions like "show me ADU permits in Oakland over $200K this year."
Thirty-nine people, headquartered in Lafayette but spread across time zones. The about-page photos lean less startup-poster and more group-text: team hikes, paddleboards, dinners, the Golden Gate Bridge in the background. There is no foosball table on display, which is probably also the point.
The engineering stack reads like a curated tasting menu of modern data tooling - dbt, DuckDB, Prefect, Snowflake, Athena, Terraform, Docker, Fargate. The use of Claude alongside scikit-learn suggests a pragmatic AI posture: classical ML where it works, LLMs where they earn their keep.
Total funding crosses $6M. Capital earmarked for ingestion, team, and expansion beyond permits into adjacent municipal datasets.
A natural-language interface sits on top of the permit graph. Customers can ask questions instead of writing SQL.
A command-line tool optimized for LLM agents - one of the first products of its kind to be designed AI-native from the outset.
Three months after that Tuesday-morning upload, the homeowner whose roof inspired the permit gets a knock at the door. It is a sales rep for a solar installer who knew - somehow - that the house was a candidate. The pitch lands. The panels go up. The next permit gets filed. The pipeline picks it up. Ninety seconds later, a battery-storage company in Denver gets a row in its dashboard.
The clerk in Kern County still has no idea. She is, by now, on her four hundredth scan of the morning. But the file she uploaded is no longer a file in a folder. It is a node in a graph that 39 people in Lafayette built, on purpose, because they decided that paperwork is a feature.
Shovels did not invent the building permit. It just made it readable.