Domain:
Personal Project | Fintech | Tax & Fiscal Tools
Context:
Freelance finance management tool
Role:
Solo Designer & Builder
Method:
UX | Interaction Design | AI-assisted workflows | Vibe coding | Claude Code | Figma | Perplexity | Gemini API | Supabase
Fiscal Copilot for Freelancers
about.
Anyone earning freelance income on the side of a full-time job has to track client invoices and documents, stay ahead of fiscal deadlines, and watch thresholds that change their tax obligations, all while that information is spread across multiple government platforms.
Existing tools do not fit exactly the problem, as invoicing software is built for businesses and focused on accounting, project management tools are too generic. I built a platform focused on the freelancer's side of the equation, shaped by a problem I ran into myself. Built as part of the Vibe Coding for Product Designers and PMs course at TheStarter.
approach.
This project doubled as an experiment in AI-assisted building, with a heavy focus on using LLMs for research, ideation, execution and learning along the way.
I used Claude Code to build the application end-to-end: frontend, database, and API integrations, guided by a written specification of features, flows, and decisions I created upfront. Google Gemini's API powers the in-app AI: extracting real data from uploaded invoices and documents, so users don't have to enter everything manually.
Building with Claude Code on VS Code > the commit history on the left shows the iterative build process over several weeks.
key decisions.
Research (via Perplexity) confirmed a gap: existing invoicing tools are built for full-time freelancers or small businesses, with too many features, while project management platforms are too broad for someone managing a full-time job and occasional freelance income.
I scoped V1 around this specific niche, designing a client and document hub, a fiscal calendar for Portuguese deadlines, and automatic VAT/Social Security threshold tracking based on real invoice values.
For the technical stack, I chose Supabase for the database and Gemini's API for document data extraction, as both are robust, well-documented, and free at this scale, well-suited for a solo-built project, with room to grow toward AI-generated fiscal recommendations, document templates, or other integrations.
result.
The result is a working tool, live and testable today: a client and document hub, a fiscal calendar that flags upcoming deadlines, and automatic VAT and Social Security threshold tracking based on real invoice data, which is extracted using AI.
Building it end-to-end, from research through design to development, gave me hands-on experience with AI-assisted workflows beyond design, and the confidence to use these tools on future projects.
The next step is testing with real freelancers to see where it holds up and where it needs to grow, and validating if the niche I identified through research actually holds up in practice.
learnings.
Building this gave me hands-on confidence using LLMs beyond research and ideation > for development, database work, and API integration, taking the project further than originally planned. Looking ahead, I'd focus on a more robust component library, deeper accessibility improvements, and testing with real freelancers before adding new features.
testimonial.
Ariana took a real problem faced by freelancers in Portugal — managing invoicing, VAT limits, and Social Security — and transformed it into a clear and actionable dashboard. Throughout the course she showed great initiative, autonomy in exploring AI tools, and a constant focus on user experience. The result is a 'Fiscal Copilot' prototype that is not only well-designed, but also demonstrates very well her ability to connect technology to real people's problems.
André F. Costa
Instructor at The Starter
