UX analysis was a bottleneck at the agency. Every project required manually pulling data from GA4, Search Console, Optimal Workshop, and user interview transcripts then synthesizing everything into a coherent report. The process took hours per project, outputs were inconsistent between analysts, and quality depended entirely on how much time was available. There was no standardised way to ensure every client got the same depth of analysis.

The bottleneck wasn't data collection — it was interpretation. Taking disparate sources and finding the patterns is exactly what large language models are built for. If the model could be given structured inputs from each source and instructed to cross-reference them systematically, it could produce a first-pass analysis in seconds that would previously take hours.

A full-stack UX analysis platform with four core capabilities:

• Report Generation Upload GA4 exports (PDF or screenshots), Search Console data, and user testing transcripts. The Claude API cross-references all inputs simultaneously to produce a structured report scored across four dimensions: UX Design, Performance, SEO, and Analytics Health.
• Live Data Connection Connect directly to GA4 via Property ID and Search Console via site URL for real-time data pulls without manual file exports. Configurable date ranges up to 90 days.
• Client Management & Trend Tracking Organize analyses by client with full historical tracking. When a client has been analysed before, the tool automatically detects trends and changes between reports — turning one-off snapshots into ongoing performance monitoring.
• Prioritised Action Plan Every report concludes with a ranked action list scored by effort and impact, with quick win flags for items completable in under a day. Output is formatted and ready to share with clients directly.

What previously took several hours of manual synthesis now takes minutes. Every project gets the same structured analysis regardless of time pressure. The AAJ and NAPFA case studies on this portfolio both benefited from outputs generated through this application.

Email signatures sound simple until you have multiple departments with different requirements. Sales signatures need a personalised meeting scheduler link and every salesperson's link is different. Other departments have their own required elements and formatting standards. Managing this manually meant inconsistent signatures across the agency, repeated requests to update individual links, and no standardised way to onboard new team members with a correct signature on day one.

The problem wasn't one signature — it was a signature system with conditional logic. AI could handle the output formatting and ensure consistency across every variation automatically, surfacing only the fields relevant to each department.

A web application where employees upload a logo, select their operating system, and enter their details. The tool surfaces only the relevant fields per role, produces a correctly formatted HTML email signature, and includes step-by-step Outlook upload instructions — removing the need for any manual formatting or IT involvement. New team members can generate a compliant signature independently on their first day.

Eliminated a recurring manual process across the entire agency. Signature consistency improved immediately — no more outdated meeting links or incorrectly formatted signatures. The live version is accessible via the link on the left.

Most people walk into a mechanic with no idea what a diagnostic code means, what a repair should cost, or whether the recommended work is even necessary. This power imbalance costs consumers billions annually in unnecessary repairs and inflated pricing. The information exists — it's just scattered across technical databases, forums, and manufacturer documentation that the average person can't navigate under pressure.

This isn't a mechanical problem — it's an information access problem. Consumers don't need to become mechanics. They need enough context to have an informed conversation with one, walk in knowing fair pricing, and leave knowing whether they got a good deal. The goal was to build the tool I wish existed every time I've had car trouble.

A React application connected to live automotive APIs across seven core features:

• Repair Analysis Enter a diagnostic code to get a plain-English explanation, a severity verdict, cost estimates from multiple sources (RepairPal, AAA, dealer average, independent shop), a symptoms tab, visual guide, and proof checklist. Claude AI powers the analysis and verdict.
• Vehicle Value Estimate Shows current vehicle value, post-repair value, and value lost to existing issues so users can make informed decisions about whether a repair is worth it.
• VIN Lookup Search by 17-character VIN to pull the vehicle's full profile, saving it to a personal garage for future sessions.
• Maintenance Tracker Tracks service history per vehicle with status indicators for overdue, up to date, and unknown items across oil changes, tire rotation, brake inspection, filters, and more. Logs service records with mileage and date.
• Local Deals & Gas Prices Surfaces real shop deals near the user's location filterable by category (oil change, tires, brakes, car wash, inspection) with distance, savings, and deal status. Live gas prices by zip code with monthly and yearly fuel cost estimates and gas cashback recommendations.
• Safety Recalls Pulls active NHTSA safety recalls for the user's registered vehicles automatically.

Currently in final development ahead of public launch. All features are built on live APIs with real data — not mock data or placeholders.