Founder · Product Architect · Software Engineer
In development · Testing
FillMyDocs turns form filling into a guided conversation. It detects what the document needs, preserves information that is already filled, handles both interactive and flat PDFs, and lets the user review the result before downloading it.
Overview
PDF forms vary wildly: some have real fields, some are flat or scanned, some are partially completed, and many render differently across devices.
FillMyDocs analyzes the uploaded PDF, identifies what information is missing and turns technical field names or blank areas into simple questions. The user answers in everyday language while the system prepares the actual document in the background.
The implementation supports Hebrew and English, interactive PDF forms and vision-based detection for flat documents, live preview, selective refill of existing values, placement correction and final download.
Product scope
The user sees three simple steps, while the system handles several document-specific problems behind the scenes.
The document enters one guided flow instead of sending the user into a generic PDF editor.
The system can identify the form and explain what it is for before asking the user to complete it.
Interactive fields are read directly; flat or scanned pages can fall back to visual field detection.
Technical PDF fields are translated into short localized questions so the user only needs to provide the missing information.
Already-filled values can stay untouched, while the user can explicitly choose a field they want to replace.
The form is rendered on the server and shown as consistent page images across desktop and mobile.
When visual detection is approximate, the user can adjust the detected area before approving the final result.
The completed document can be reviewed, corrected if necessary and downloaded only when the user is satisfied.
Users & use cases
The main user is the person completing the form. The product adapts to how the document arrives instead of forcing every PDF through the same path.
The system identifies the fields that need information and turns them into a guided sequence of simple questions.
Existing values stay in place by default, so the user answers only what is missing or deliberately chooses a field to refill.
When real form fields do not exist, visual detection finds writable areas and the user can verify placement before approving the result.
Core workflow
The pipeline chooses the simplest reliable path for the document instead of treating every PDF the same way.
Architecture
The application detects what kind of PDF it received and routes it through the appropriate document pipeline while keeping the conversation and preview flow consistent.
The AI-assisted steps identify and interpret the document, while the final PDF is produced through deterministic PDF operations and remains reviewable by the user.
Engineering decisions
The most important engineering decisions came from handling the differences between real-world PDF documents and devices.
When real form fields exist, use them. When they do not, fall back to page-level visual detection.
The product first looks for structured AcroForm fields. Flat or scanned documents use a visual path that detects writable areas and later draws the user answers into those positions. Both paths feed the same conversational product flow.
Browser PDF rendering was not reliable enough for legacy and Hebrew-heavy forms across devices.
The application renders PDF pages with PDFium on the server and sends page images to the client. That gives desktop and mobile the same visual output and removes browser-specific font rendering from the preview path.
AI can suggest where text belongs, but an approximate box should not become an irreversible decision.
Vision-detected fields carry editable geometry. Users can move or resize a field when needed, and later refinement respects those manual corrections instead of silently replacing them.
A partially completed form should not force the user to re-enter or overwrite information that is already there.
Detected fields track whether they already contain content. The question flow skips those values unless the user explicitly chooses to refill a field.
A Hebrew UI is not enough if the generated document cannot render Hebrew text correctly.
The fill pipeline embeds a Hebrew-capable font, handles right-to-left text alignment and converts field geometry between the coordinate systems used by vision and PDF drawing.
A technically successful fill can still be wrong if the answer sits on a label or outside the intended blank.
The completed document can be rendered again and inspected for clear placement problems. Flagged fields can return to a focused correction flow before approval.
Engineering evidence
The codebase is intentionally small, so quality is shown through domain-specific verification rather than large repository counts.
Current product shape
Technology
The stack combines web application development, AI-assisted document understanding and low-level PDF processing.
Product principle
FillMyDocs hides the differences between interactive fields, scanned pages, coordinate systems and font rendering behind one simple flow: upload, answer, review and download.