Opening day of Compass Coffee in the Shaw neighborhood of D.C. on Monday, September 22, 2014. (Melina Mara/The Washington Post)

At Compass Coffee, founders Harrison Suarez and Michael Haft try to ensure that each cup has the same medley of flavors — with hints of vanilla, perhaps, or notes of caramel — as the last. The two former Marines are constantly looking for technology to make the business’s operations more consistent.

The temperature inside the shop, in Northwest Washington’s Shaw neighborhood, is modulated by a WiFi-enabled Nest thermostat that adjusts settings to users’ preferences over time. Suarez and Haft designed the chairs, tables and shelves with SketchUp, a 3-D modeling program. If they need to replace a chair, they can replicate it according to the exact specifications.

And with an experimental software system designed in-house, the Compass team is attempting to apply precision to the roasting process while maintaining a personal touch. Almost since the shop’s opening in late September, the company’s small roaster, used for samples, has been transmitting data about each test batch to the Internet, where Haft, Suarez and head roaster Brandon Warner study the details before roasting a larger, production batch.

Compass isn’t the first coffee business to embrace technology. Last year, Starbucks, for instance, announced plans to double the number of Clover coffee-brewing machines in its stores. When connected to the Internet, the machines can brew coffee according to customers’ individual preferences, Bloomberg News reported.

Compass occupies a converted laundromat in Shaw. It has large sidewalk-facing windows, and there are tables arranged neatly throughout the shop. A counter with bar stools offers a clear view of the roasting area, so customers can watch while drinking their coffee. Sometimes Haft and Suarez invite them to smell a sample of beans.

On Opening day of Compass Coffee barista Brittany Pugh makes and serves a latte. (Melina Mara/The Washington Post)

Before serving customers coffee brewed from an unfamiliar bean, Haft, Suarez and Warner do several rounds of taste tests. Their roaster for samples, manufactured by Coffee Per, can handle a pound of beans. It retails online for about $9,000.

Haft installed a USB port and a digital thermometer in the samples roaster, which automatically transmits data points about each test roast — temperature changes and moisture levels, for instance — to open-source software called Artisan.

If the three-member team likes a type of bean and the way it was roasted, Warner retrieves the data from Artisan, and applies the settings to a larger, industrial-size roaster to reproduce the results for large numbers of customers. If the three change their minds about the taste, they adjust a variable and try again.

“Bringing the temperature up to point X, at time Y, [might] give us a little bit of strawberry notes or the chocolate we wanted,” Warner said. “We take that information and go back to the drawing board a little,” Warner said.

Artisan records “roast profiles” for each batch, including graphs that track time and temperature. The team also enters the ratio of bean types in a blend, so the proportions can be re-created later, along with any comments. After a few trials, the three usually agree on a recipe. Warner then sets the industrial roaster based on the coffee’s Artisan profile.

The industrial roaster, a Kestrel, made by a California company called Loring, cost about $150,000, including installation, and can roast about 70 pounds of coffee beans in 15 minutes. It is also WiFi-enabled and sends Warner’s laptop a roast profile after each batch.

In the very beginning, before the small roaster was connected to the Internet, the team measured and recorded data points by hand, entering the information into an Excel spreadsheet, Suarez said. But the three tended to record different data points, depending on what each found significant, making batch-to-batch comparisons difficult.

Opening day of Compass Coffee in the Shaw neighborhood of D.C. (Melina Mara/The Washington Post)

“We realized we needed to come up with standardized way of collecting all this information so we could go back” and re-create it, Suarez said.

With the company’s system, the three can “ ‘copy and paste’ [settings] every time, and we’re able to hit those same marks without relying on note-taking,” Haft said. “When we finally have a coffee we want, or a blend we’re really proud of, it’s very easy to scale up.”

The system also enables the company to deliver consistent roasts to other businesses that prepare and sell Compass Coffee. If Dog Tag Bakery — a nonprofit store in Georgetown operated by wounded veterans and their spouses — wants its Compass Coffee “to taste like a ‘Dog Tag’ blend . . . we have a consistent input of the coffee and a consistent roasting style,” Haft said.

Parts of the brewing process are also automated. Taps for brewing pour-over coffee, which are near the counter, are controlled by a programmable system called Modbar that lets Haft and Suarez set the temperature and pressure of the water.

Haft and Suarez plan to continue experimenting with and adjusting the selection of the coffee they serve. When taking notes on a sample, they often make use of an aroma kit with vials of scents, such as clove and vanilla, to make sure that they are applying the same flavor benchmarks.

They are hesitant about adding more technology to the roasting and brewing processes.

“The computer is better at modulating temperature than we are,” Haft said. “But the computer can’t smell [the beans] the way Brandon can smell them.”

And even though the automated roaster settings can make the end product more consistent, Warner said, the organic makeup of the raw beans still changes from batch to batch.

“We’re dealing with non-stable agriculture in green coffee,” Warner said. Two shipments of beans can be from the same farm and one can be slightly denser or a slightly different color than the other.

“Every single batch is still measured,” he said. “Every single batch is still quantified. . . . Then we have to make adjustments on the profile that we use.”