The Situation
Every stormwater report CityScape writes starts with the same groundwork. What soil is on the site, and how well does it drain? Is it in a FEMA flood zone, a wetland, or Maryland’s Chesapeake Bay Critical Area? Which watershed does it drain to? Who owns the parcel, and how is it zoned?
None of that is secret. It’s all public. But it’s scattered across separate federal, state, and city maps and databases, each with its own interface and its own way of making you search. Pulling it together for a single site used to take a principal civil engineer a full workday.
So the most experienced people at the firm were spending a day as very skilled copy-and-paste machines before the real engineering could even start, and doing it again on every new project.
The goal: Significantly reduce the amount of time needed to complete a report that is required by the state in order to submit plans.
What was built
A site screening tool where the engineer draws a shape around the property and gets every piece of site data at once, plus a one-code handoff that sends it straight into CityScape’s Stormwater Assistant in Copilot to write the report sections.
- Draw the site instead of searching for it — look up the address, trace the property boundary on an aerial map, and click Analyze Site. The tool checks that exact shape against every data source at once.
- Every layer the report needs, in one pass — soil map units with the share of the site each one covers, hydrologic soil group and hydric soils, FEMA flood zones, Maryland Critical Area, wetlands, nearby streams, watershed and HUC code, plus Baltimore City parcel and zoning data.
- A map the engineer can check by eye — switch between aerial and street views and toggle each data layer on or off to confirm what the numbers say.
- A Site Intelligence Record ID — one click saves that site’s results and generates a code to go with them.
- A direct line into Copilot — paste the code into the Stormwater Assistant, the agent CityScape already relies on, and it pulls the full record and builds the project data sheet tables and write-up, ready to paste into the report.
What the engineer does for each new site:
- Search the address and draw a polygon around the site.
- Click Analyze Site and check the results and map layers.
- Click Get Record ID and copy the code.
- Paste the code into the Stormwater Assistant in Copilot.
- Paste the tables and text it generates into the stormwater report.
Draw a shape, copy a code, paste the answer. The hours of digging in between are gone.
How It Was Done
The screening tool is a lightweight web app on Cloudflare Workers, so it opens in any browser with no login and nothing to install. When the engineer draws a site, the tool sends that exact boundary to multiple public data services (soil survey, FEMA, Maryland Critical Area, the National Wetlands Inventory, and Baltimore City parcel and zoning data) and combines the answers into one set of results.
The harder problem was getting that data into Copilot without anyone retyping it. Each analysis is saved on Cloudflare under its own Site Intelligence Record ID. The Stormwater Assistant was extended so that when someone gives it a record ID, it retrieves that site’s data and turns it into the tables and text the report needs, working from the record rather than the open web.
Total active build time: ~16 hours + testing time against real CityScape project sites.
The Outcome
What used to cost a principal engineer a full workday now takes minutes, and it runs the same way every time. The site data comes straight from the source instead of someone’s notes, arrives formatted for the report, and frees up the engineer’s day for the judgment calls only an engineer can make.
This is the first step, not the finish line. The next phase automates more of the stormwater report itself, with the goal of getting the whole thing down to an hour or less of engineer time.
Pulling site data for a stormwater report used to take me or one of my senior engineers a full day — bouncing between soil surveys, FEMA maps, Critical Area maps, and city parcel records, then retyping all of it into the report. Now I draw a box around the site, paste one code into Copilot, and the tables and draft narratives are ready to drop in.
What Made this project work
- The data was already public and reliable, just scattered. The problem was never finding the information, only gathering it, and gathering is exactly the kind of work a tool should take over.
- It built on something the team already trusted. The handoff plugs into the Stormwater Assistant the engineers were already using, instead of adding one more AI tool to learn.
- The engineer stays in charge. The tool gathers and formats, and the engineer still reviews every value and decides what goes in the report.
The most expensive hours in a small firm often go to work that isn’t the job: gathering, reformatting, retyping. Look for the places where your most senior person is doing a junior person’s research. That’s usually where the fastest win is.