PDF Penguin

While developing my cooking assistant app, Chefie, I needed a way to extract structured ingredient and nutrition data from USDA PDFs. The datasets were available, but they were formatted as complex, unstructured PDFs that were difficult to work with programmatically. Existing tools were unreliable or too technical, so I built a clean, AI-powered tool for parsing and exporting PDF data into usable JSON.

PDF Penguin has since evolved into a standalone product with broader application across document-heavy industries.

My Role

This was a completely solo project, where I handled every aspect from product vision to deployment.

• Founder
• Product Designer (UX & UI)
• Frontend Developer (React, TailwindCSS)
• AI Integration (Vision Models)

Step 1: Empathize & Research

I explored a range of existing PDF parsing tools including Tabula, Adobe Acrobat's OCR export, and DocParser. While all three were technically capable, they presented significant barriers for non-technical users — requiring either installation, rule-building, or an understanding of export settings and schemas. I tested each one by attempting to extract structured data without relying on documentation or setup guides, simulating the experience of a first-time user with minimal technical background.

Across the board, I encountered slow onboarding, confusing interfaces, and results that required multiple adjustments or retries. Even simple use cases like "extract table data" demanded upfront learning or configuration. These pain points reinforced a clear gap in the space: a need for a tool that provides structure, flexibility, and results — without setup or specialized knowledge.

The table below summarizes how PDF Penguin compares to these tools, based on that usability-first evaluation.

Step 2: Define

The research made one thing clear: the biggest barrier wasn't technical capability — it was usability. Most tools assumed the user had experience with templates, schemas, or parsing rules. I defined the core product need as creating a parsing tool that eliminated setup entirely. PDF Penguin would focus on a single principle: let users describe what they want in plain language, and deliver results instantly — simple to use, with no learning curve.

Step 3: Ideate

I had a clear mental model of how the product should behave, so I quickly sketched a basic 2-panel layout idea: Upload (left) → Output (right), supported by a flexible prompt box to direct the AI. The goal was instant clarity, minimal onboarding, and the ability to adjust the output on the fly.

Step 4: Prototype & Design

With a clear layout already in mind, I moved directly into high-fidelity interface development using Vercel. Rather than spending time on static design tools, I focused on live iteration — adjusting spacing, labels, and user flows in context. This allowed me to make real-time decisions based on how the interface behaved, rather than just how it looked. Cursor supported this process by helping structure responsive components and quickly refine interactive behaviors.

Step 5: Test & Iterate

Each time I implemented a new UI or prompt behavior, I tested it by uploading different document types — invoices, receipts, reports — and refining the prompt UX to guide the AI parser. After discovering poor output from vague prompts, I added a customizable instruction field and clarified the placeholder text to guide user input.

Once the interface flow and AI behavior were reliable, I built the full app in Cursor with a React + TailwindCSS frontend, integrated OCR and OpenAI APIs, and deployed it via Vercel. The result is a working product with real users, capable of turning even messy PDFs into structured data in seconds.