We replaced the AI model, rebuilt the processing pipeline, and shipped improvements to how we handle complex form layouts. Because of these changes, we recommend re-uploading your forms and filling them out again to see the improvements in action. Below is a quick summary of what changed and how it improves your results.
New AI model. Smarter parsing. Faster throughput.
9 improvements across reading, structure, and precision – all live, all automatic.
Smarter reading – The AI now reads and understands your documents better
1. GPT-5.2 across the entire platform
We replaced the previous-generation model with OpenAI’s latest for both text extraction and field filling. We tested multiple variants (GPT-5, 5.1, 5.2, nano) and selected the best one for each type of operation.
2. Page-by-page source reading
Every page of your source documents is now extracted individually. Even 70+ page PDFs are fully read, so no relevant data is missed.
3. Intelligent relevance mapping
For large source files, the system identifies which sections are relevant to each form field instead of processing everything at once. This reduces errors and speeds up filling.
Better structure understanding – The AI now sees how your forms are organized
4. Cross-page field dependencies
A checkbox on page 5 that says “Same as mailing address” (entered on page 1). A total on page 8 that sums page 6. The AI now detects these relationships during fine-tuning and fills them out consistently.
5. Section-based processing
The system now detects single-column vs. multi-column layouts and groups related fields into logical sections that can span page breaks. A single form can use different processing strategies for different pages, selected automatically.
6. Row-by-row table filling
Tables are now filled one row at a time, with values mapped to the correct columns based on headers. This eliminates misplaced data in wide tables with many columns.
Precision and speed – Fields that fit, text that sounds like you, faster throughput
7. Smart text overflow handling
The system now checks every filled field against its actual pixel boundaries. When text does not fit, a separate AI operation reformats or shortens the content to stay within bounds without losing meaning.
8. Style replication for generated text
For narrative fields (descriptions, clinical notes, progress reports), the AI matches the writing style from the examples you provide. The filled form reads like it was written by you.
9. Rebuilt processing pipeline
A new task queue system supports up to 30 parallel operations. Forms with multiple sources or complex layouts process faster, and batch fills benefit from the increased throughput.
What to do next
These improvements apply automatically to every new filled form. If your forms were fine-tuned more than a few weeks ago, regenerate fine-tuning to apply the latest AI improvements. Your manual edits are always preserved.
Works with all your existing forms
All existing forms have been migrated to support the new improvements. You don’t need to re-upload anything. Regenerating fine-tuning is optional but recommended for forms that rely on complex layouts or large source documents.
If you have questions about any of these improvements or want to discuss how they apply to your forms, reach out at [email protected].