Turns matter data into court-ready form packets, reviewed and filed from one place.
DraftEngine is Gemini Legal’s platform for preparing and filing court form packets: a B2B enterprise SaaS product serving everyone from solo practitioners to law firms with 500+ attorneys. As the sole and founding product designer, I designed it end to end, from defining the problem space to the shipped product. This case study is about the design decision that shaped the product’s direction the most, and how I pivoted when the first round of testing proved my brief wrong.
Legal professionals spend hours manually filling court forms across multiple platforms, leading to wasted time and errors.
DraftEngine auto-fills forms with CMS data and centralizes filing and case management, with integrations, alerts, and dashboards to keep everything on track.
Streamlined the entire form-filling process by consolidating form completion, team management, and court interactions into a single platform, eliminating the need for manual emails and external websites.
Every matter, packet, and form page organized in one expandable grid, with status, deadlines, and quick actions always within reach.
Surfaced every case touchpoint, from deadlines and team notes to in-progress and completed packets, in a single hub, with automatic highlighting for items due within seven days.
Court forms are automatically pulled and pre-filled with case data, so fields are always up to date and no one has to manually track changes or re-enter information.
Users can edit any field directly on the form and instantly see whether the data was pulled from the case management system or entered by hand, so there is no guessing where information came from.
Behind the scenes of how DraftEngine came to be, from the first brief to the shipped product, including the moment the research sent me back to the drawing board.
Legal professionals spend hours filling court forms by hand, and every hour of re-typing is another chance for an error to reach a judge. At a high level, they are assigned cases, work through the required forms for each case, and submit completed forms to the court.
Receive cases/matters → Fill out forms → Submit to court
The high-level workflow sounds simple, but the reality is fragmented. Filing a single matter meant juggling three or four tools, from the CMS and a document editor to the court portal and more, with attorneys re-entering the same data at every handoff and no one owning the full picture. The real burden wasn’t just the repetitive typing but the cognitive load of tracking what was current across systems.
Current workflow
To solve this problem, I first conducted user research in several formats with legal professionals to deeply understand their workflows and identify key pain points. I paired interviews with contextual inquiries on purpose. Attorneys describe their workflow as three clean steps, but watching them work exposed the hand-offs between systems where time and accuracy actually leak. The internal survey added the view from the people who work alongside attorneys every day.
How might we let attorneys move faster from matter data to filed document without re-entering information or switching platforms?
The new workflow flipped the script. Firms already had the matter data in their CMS, so DraftEngine pulls it in once, applies it across every form in a packet, and produces filings. Same data, filled correctly, every time. This was the hypothesis we took into design, and every later decision was measured against this flow.
Before committing to a layout I went wide: twenty directions for what the working surface could be, then prioritized features based on user value and insights from competitive research. Sketching the extremes made the trade-offs visible before any of them were expensive to build.
The first version bet on speed. Attorneys land on a dashboard showing the day’s priority tasks, then click into one to open a packet, a group of forms pre-assembled by the firm’s paralegals. The review screen pairs all forms in the packet on the left with matter data pulled from the CMS on the right, letting attorneys scan, edit, and fill any gaps before approving the packet for filing. It did exactly what the research brief asked for.
Then user testing with eight participants told two different stories.
While the platform made filling forms significantly faster, very few participants said they would actually adopt it. When I pushed further in follow-up conversations, the answer surfaced. The blocker wasn’t speed. It was reliability. We had solved speed and we hadn’t solved trust. Attorneys stake their license on every filing, and a fast tool they can’t verify is a tool they won’t use. Rather than tune the dashboard, I rewrote the question and took the redesign back to the team with the test results as the argument.
How might we let attorneys move faster from matter data to filed document without re-entering information or switching platforms?
How might we help attorneys confidently stake their license on automated work?
I went back to the testing feedback and broke “trust” into something I could design around: a named framework with three pillars.
Can users tell which data is pulled versus entered, and whether it’s up to date?
Are users confident nothing is missing or urgent?
Does the automation work for users, or decide for them?
Designing around accuracy meant removing any reason to double-check the automation.
Designing for completion meant making “done” unambiguous.
Designing for control meant adding leverage without taking any away.
After redesigning around trust, users filled out forms even faster thanks to clearer information organization, and every participant said they would adopt the platform because of its reliability.
The redesign kept the speed gain and turned adoption from two participants to all eight. DraftEngine shipped with the trust framework at its center.
Beyond the test room, with legal professionals using DraftEngine in their day-to-day work:
The problem might not be the problem.
Our brief was “make form-filling faster,” but the real problem was trust. First-order briefs describe the pain. Second-order research reveals the job.
Testing is a framing tool, not a validation tool.
Most teams use usability testing to confirm a design works. This project taught me it’s better used to pressure-test whether the problem was framed correctly at all. Round 1 didn’t tell me the buttons were wrong. It told me the problem statement was. The best testing shifts the question, not just the answer.
Name the framework so it travels.
Once we broke trust into three dimensions, the team had a shared vocabulary we reached for in every review that followed. Screens get redesigned, but frameworks stick.
I established a modular design system in Figma from the ground up, leveraging variables and token architecture to enable long-term scalability. Throughout development, I collaborated with engineers to implement components accurately and maintain cross-platform consistency.
Used Figma’s variable modes to create themes systematically.
All icons map directly to their FontAwesome sources.
QA-ed the code base to make sure the naming and values were implemented correctly.
I then transferred the Figma assets into a design repo using Claude Code and set up the design system in Claude, so I could use it for AI prototyping. I started using it to rapidly prototype concepts to showcase to the team during my design process.
I also ran multiple accessibility checks to ensure the design is inclusive, usable, and meets accessibility standards across various scenarios.