Case Study
Personal Project
Pin.point
A Figma plugin that runs UI/UX audits powered by the Claude API — and writes the findings directly onto the canvas.
The Problem
Audit output lives outside the file. The loop is broken by default.
Most audit tools produce output that lives outside the file. PDFs. Spreadsheets. External reports. The designer reads the findings in one context, then switches back to Figma to apply them. Every context switch costs time and dilutes the connection between the finding and the element it refers to.
Manual audits have a second problem: consistency. Applying ten frameworks to a design by hand produces results that vary depending on how much time you have, how familiar the frameworks are, and what you happen to notice on that day. The same screen audited twice rarely produces the same list.
"Most audit tools produce output that lives outside the file. PDFs. Spreadsheets. External reports. The loop is broken by default."
The core problem Pinpoint was built to solve.
The Decision to Build
From a prompt in Claude to a plugin inside Figma
The starting point was a structured prompt in Claude — a system designed to evaluate designs against UI/UX frameworks and return a scored, categorized list of findings. It worked. But the output lived in a chat window. To act on a finding, I had to remember it, switch to Figma, find the element, and apply the change. The friction was real.
The decision to build a Figma plugin came from one question: what if the audit happened where the work happens? The prompt system I had already designed became the engine. The plugin became the interface.
The choice of Claude specifically was deliberate. The evaluation required not just pattern recognition but contextual judgment — which frameworks apply to this type of interface, how severe is this finding relative to the others, what is the realistic effort to fix it. That level of reasoning required a model capable of nuanced analysis, not just classification.
How It Works
Three steps from selection to findings on the canvas
1
Configure
Select a frame in Figma and open Pinpoint. Choose the audit mode — UI only, UX only, or Full — the WCAG accessibility level, and the page scope. One click: Run audit.
2
Analyze
Pinpoint sends the selected frame to the Claude API with a structured instruction system. Claude evaluates the design against the applicable frameworks, scores each finding by severity and estimated effort, and returns a structured set of results. The process takes approximately one minute depending on file complexity.
3
Results on the canvas
Findings appear as annotation cards written directly onto the Figma canvas, with arrows pointing to the specific elements they reference. The plugin generates a score summary — UI score, UX score, and a combined full score on a 1–10 scale. From the same interface, the designer can re-audit after making changes, translate the findings, or export the full HTML report.
Plugin running a Full audit on the Banco do Brasil case. 30 issues · 7 sections · Frame 1608.
Processing state — Claude API evaluating the selected frame.
The Output
Two artifacts. Two audiences. One workflow.
Pinpoint produces two distinct outputs designed for two distinct purposes.
Canvas annotations
For the designer
Each finding is placed directly on the canvas next to the element it refers to, categorized by layer (UI or UX), labeled by severity, and linked to the framework that flagged it. The designer never has to leave the file to understand what needs fixing or why.
HTML report
For stakeholders
Organizes findings into a priority matrix (Quick Wins, Address, Later), an Impact × Effort scatter plot, a methodology section, and a filterable findings list. Designed to support design decisions in presentations and reviews — turning an audit into a defensible artifact.
The Scoring System
1–10. Transparent formula. Documented in every report.
Each audit produces three scores on a 1–10 scale: UI, UX, and a combined full score. The formula weighs findings by severity — high findings carry the most weight, medium a moderate weight, low findings minimal weight — and normalizes the result by the number of sections audited. WCAG violations receive additional weight because they represent real user impact, not design preference.
2.3
UI Score
3.5
Full Score
5.4
UX Score
Example from the Banco do Brasil audit. The combined score weights UI at 60% and UX at 40% — UI findings are more objectively measurable, while UX findings involve more interpretive judgment.
The scoring system was designed with Claude, refined through use across four projects, and documented in the HTML report so any stakeholder can understand exactly how the number was produced.
The Frameworks
Contextual selection, not blanket application
Applying all frameworks to every project generates irrelevant findings. A consumer mobile app and a financial enterprise dashboard have different failure modes. Pinpoint selects frameworks contextually based on the type of interface being audited.
Always applied
Refactoring UI
Any visual interface — detects size-only hierarchy, arbitrary spacing, flat design without depth.
WCAG 2.1
Always applied — impacts real users. Minimum contrast 4.5:1 text, 3:1 UI, touch targets ≥44px.
Bringhurst
Typography exists in every interface. Detects scale without progression, inadequate line-height, weight without hierarchy.
Applied by context
IBM Carbon
Financial and enterprise products with complex components and high density. Detects missing states, grid inconsistencies.
Material Design 3
Reference for colour systems and elevation in cross-platform web products.
UX frameworks
Nielsen Norman Group
Usability heuristics — visibility, feedback, error prevention.
IDEO / Design Thinking
User need alignment, empathy gaps, solution–problem fit.
IxDF
Interaction patterns, mental models, affordance failures.
Gestalt
Visual grouping failures, figure–ground issues, proximity and similarity violations.
Shneiderman's 8 Golden Rules
Consistency, shortcuts, closure, error handling and reversal of actions.
Don Norman — Emotional Design
Visceral, behavioural and reflective design alignment.
Laws of UX
Fitts's Law, Hick's Law, Miller's Law, Jakob's Law violations.
Evolution
From a prompt in Claude to a versioned plugin
v4.4 current
Pinpoint reached v4.4 through sustained iteration across four projects and months of use. The core idea was stable from the beginning — audit inside the file, results on the canvas — but the execution evolved significantly.
Early versions produced results as plain text annotations. Later versions introduced structured cards, severity labels, and the arrow system connecting findings to specific elements. The HTML report started as a flat list and grew into a navigable document with a priority matrix, an Impact × Effort scatter plot, a methodology section, and a full framework documentation layer.
Each version was tested on a real project. The gaps that surfaced in use became the next iteration.
In Practice
Four projects. One consistent finding: building and auditing are different modes.
Pinpoint has been used across four personal projects. The most instructive was a retrospective audit of the Contabilizei Onboarding 2.0 — a project I had designed and delivered from scratch.
Designing a project and auditing it are genuinely different cognitive modes. Building requires forward momentum; auditing requires critical distance. Running the Onboarding through Pinpoint surfaced information density issues and content hierarchy problems that weren't visible during the creation process — not because the original work was careless, but because the evaluation framework was different.
That gap — between what a designer sees while building and what an audit reveals after the fact — is the core value Pinpoint provides.
Contabilizei Onboarding 2.0
Retrospective audit · surfaced density and hierarchy issues
Banco do Brasil portal
30 issues · 7 sections · UI 2.3 / UX 5.4 / Full 3.5
Personal project #3
Full audit mode
Personal project #4
Full audit mode
Reflection
What it does well. What it doesn't yet do.
Pinpoint is a personal tool, not a commercial product. It requires the designer to bring their own Anthropic API key, which limits distribution. It was built to solve a specific problem in a specific workflow — and it does.
Current limitations
Pinpoint doesn't yet evaluate interactive behavior, test against real user flows, or automatically compare a design against a previous version. The baseline tracking feature visible in the plugin interface is the beginning of that last capability — but it isn't complete.
What the next version closes
The next meaningful iteration would close the loop further — not just finding problems, but tracking whether they were fixed between audits and measuring the score delta over time. The goal is to make improvement visible, not just problems.