paperPilot

home page
upload your form
features and how it works
fill it by guidance
do it
save it or generate guide text

Inspiration

PaperPilot was inspired by the frustration of filling out important forms without fully understanding what was being asked. Many real-world documents—such as visa applications, college forms, and insurance paperwork—are written in dense, formal language designed for systems and reviewers, not for the people completing them.

While there are many tools that summarize PDFs or allow users to ask questions in a chat format, these tools often fall short when accuracy matters. A misunderstood field, a wrongly formatted answer, or a skipped declaration can lead to rejection or delays. This highlighted a gap: people don’t just need explanations—they need guided, step-by-step help to complete forms correctly.

PaperPilot was created to turn confusing paperwork into a clear sequence of actions, helping users understand what to fill, why it matters, and how to avoid mistakes—without sacrificing privacy.

What it does

PaperPilot is a local-first paperwork assistant that analyzes uploaded documents (PDFs, images, or DOCX files) and converts them into structured, actionable steps.

Key capabilities include:

Breaking documents into logical steps grouped by intent (identity, verification, declarations, etc.)

Highlighting risk levels for each step and explaining why errors matter

Separating fillable fields from informational or instructional content

Providing safe draft guidance using professional templates instead of fabricated data

Tracking progress so users can mark steps as completed or skipped

Exporting a filled PDF when the document contains true AcroForm fields

The result is an interface that helps users understand their paperwork before submitting it, rather than guessing or copying blindly.

How we built it

PaperPilot is built with a clear separation between analysis and presentation.

Backend

FastAPI powers the backend API

PyMuPDF is used to extract text and detect AcroForm fillable fields in PDFs

EasyOCR provides OCR fallback for scanned documents and images

python-docx handles DOCX file extraction

Custom heuristics detect field-like text and reduce OCR noise

The backend converts extracted content into a structured step format consumed by the frontend

Frontend

Built using Next.js (App Router) with React and TypeScript

Tailwind CSS and Radix UI components are used for a clean, readable interface

The main “after-upload” experience is handled by a FormProcessor component that:

Displays steps

Shows guidance and risk indicators

Allows safe draft suggestions and review

Supports filled-PDF export where applicable

The system is designed to be modular, so extraction, step generation, and UI rendering can evolve independently.

Challenges we ran into

One of the biggest challenges was handling the variability of real-world documents. Some PDFs contain clean text layers, others are scanned images, and many mix instructions, declarations, and form fields in inconsistent layouts.

OCR quality was another challenge. Scanned documents often contain noise, skew, or formatting artifacts, requiring text cleaning and heuristics to avoid misleading field detection.

Balancing automation with safety was also difficult. Automatically generating answers can be risky for paperwork, so PaperPilot had to avoid hallucinated personal data and instead provide guided templates that users explicitly review and edit.

On the frontend, maintaining readability for dense forms—without overwhelming users—required careful layout, spacing, and typography decisions.

Accomplishments that we're proud of

1.Building a system that goes beyond summarization and focuses on actionable clarity

2.Successfully detecting and filling AcroForm PDFs while maintaining user control

3.Designing a UI that separates fillable fields from explanatory content

4.Implementing a privacy-first workflow with no account requirement

5.Creating a complete end-to-end experience from upload to review to export

What we learned

This project reinforced that accuracy matters more than automation when dealing with high-stakes documents. Users benefit most when tools explain why something is required, not just what to enter.

We also learned the importance of designing AI systems with constraints. By limiting suggestions to safe templates and structured guidance, PaperPilot avoids common pitfalls of over-confident automation.

From a technical perspective, the project deepened our understanding of document parsing, OCR limitations, and building maintainable full-stack systems with clear boundaries between logic and presentation.

What's next for paperPilot

Future improvements for PaperPilot include:

Coordinate-based overlays to support filled-PDF export for scanned forms

Better form field classification using layout and semantic signals

Optional offline-only processing modes for enhanced privacy

Support for additional document types and regional form formats

More advanced review and validation checks before final submission

The long-term goal is to make PaperPilot a reliable companion for anyone dealing with important paperwork—helping them submit with confidence rather than uncertainty.