The client Stevens Industries is the largest manufacturer of commercial casework and architectural millwork in the United States. Even with a high-tech factory full of advanced machines, their quoting process was still done by hand. A small team spent hours going through architectural drawings, figuring out what was needed, and creating quotes. It took a lot of time, and mistakes were hard to catch. The Business Goal The business goal for Stevens in developing the solution with COXIT was to enhance their ability to grow by increasing the efficiency of quoting and estimating projects. Specifically, they aimed to break the traditional equation where more sales required more labor, which was becoming increasingly difficult due to labor shortages and the time needed for new employees to gain the necessary experience. By integrating technology, they sought to streamline the process of interpreting construction drawings and generating accurate estimates, ultimately allowing them to handle more projects with the same staff and improve overall productivity. Manufacturing Insights, Part 3 Once the PoC demonstrated success, the project moved into the MVP stage. This phase focused on developing a user-friendly interface and integrating essential features that would enhance the user experience. The goal was to create a functional product that could be used in real-world scenarios while continuing to build out the underlying technology.
The 4 key stages The following stages started after the testing phase. We had already build a simple demo that could read construction drawings and find the right cabinets. Now the goal was to make it work for real projects. Throughout the MVP development, user involvement was emphasized. Stevens included team members who would interact with the software daily, allowing them to provide feedback and suggestions for improvements. This engagement helped ensure that the final product would meet the actual needs of the users.
Generate Clean Output January — February 2024 Goal We wanted the system to produce clean results that could be dropped straight into a spreadsheet that Stevens could then use to import into their quoting system. That meant no fixing things by hand, no missing details, and no extra steps. What was done The system was taught to handle cabinets more accurately. If one cabinet showed up in two pieces, it combined them. It also counted the shelves to pick the right product code. The result was a clean spreadsheet, listing each cabinet once with the correct details. The drawings were cleaned up too. Markings were redrawn for easy editing, and reviewers got a helpful first draft to speed things up. Why it mattered Before this, people still had to fix split cabinets, count shelves by hand, and enter product codes themselves. After this stage, that work was done automatically — saving hours and helping the team move faster with fewer mistakes.
Add Web Access March — June 2024 Goal Make the tool easy for anyone to use, without needing help from the tech team. It had to keep drawings secure and avoid high cloud costs. What was done A simple website was built where users could drag in a file, track progress, and download results. Logins were added so people could only see their own files. Roles like Super-admin, Admin, and User helped keep things organized. To save money, the AI computer now turns on only when a file is uploaded, and shuts down when it’s done. As the tool came together, users were shown how it worked and how it could make their tasks easier and more efficient. Training sessions helped ease concerns about switching to a new system. The rollout was gradual, giving users time to adjust without feeling overwhelmed. At first, they could use the new system alongside their usual process. This helped make the transition smoother and gave everyone time to get comfortable. Why it mattered Anyone in the workflow can now use the tool on their own. Files stay secure, and the system runs quietly in the background without extra cost or support.
Improve Speed and Accuracy June 2024 — February 2025 Goal Now that the tool was in everyday use, the next step was to make scans faster and more accurate, so planners wouldn’t have to fix small mistakes or wait too long for results. What was done The system learned to read CSI/MasterFormat specification pages. Technical blocks such as Finish Schedules, Legends, and General Notes could now be read so hidden terms were easier to find. It also got better at filtering out weak matches. Processing started running in parallel, which cut wait times and reduced false positives. The exported spreadsheet was improved too. Items now show up in the right order, include counts when handling Typical situations, and pull more reliable page numbers. Behind the scenes, every scan now logs a full trace. This made it easier to spot problems if something didn’t look right. Why it mattered The scans became faster and more accurate, so staff stopped re-checking everything. And with cleaner results, the tool was ready for upcoming features like automatic depth detection where trust in the data really matters.
Prepare for Production Use Ongoing Goal Now that the system was fast and accurate, it needed to be safe and stable. Something the team could run without the fear of running into development issues. What was done All the services were connected through secure, password-protected channels. Log-ins were linked to company emails, and roles were set automatically so everyone only saw what they needed. A training pipeline was also added. When a new model finishes training, it can go live with one click and update its list of parts in the database. The system now checks each file before running. If someone uploads a scan or another unsupported format, it gives a clear warning up front. A few tricky bugs were fixed too — mainly rare cases that could freeze a job or overload the system. The AI computer still powers down when it’s not needed, keeping costs low just like before. Why it mattered These updates turned the tool into something the team could rely on every day. Drawings stayed secure, jobs ran smoothly, and new cabinet types could be added without asking software engineers to step in.
Workflow Revolution From slow, manual work to fast, automated precision — here’s how the process changed.
What Archiscan looks like today Archiscan is a website where preconstruction staff can handle blueprint tasks without extra tools or tech help. They open it in a browser, drop in a PDF, and click Process. A progress bar shows what’s happening. When it’s done, they get two files: An annotated drawing: cabinets, notes, and keywords are already marked. The markings can still be adjusted in Bluebeam if to allow for continued workflow needs. A spreadsheet: each cabinet shows up as one row, with the right product code, how many were found, and any linked notes like room names or sink types. There’s also a built-in chat box. If someone needs to check something — like “How many wall cabinets are on page A-23?” — they can ask and get an instant answer, linked to the right page.
What’s happening behind the scenes Now that the tool was in everyday use, the next step was to make scans faster and more accurate, so planners wouldn’t have to fix small mistakes or wait too long for results.
