Goril Docs

Automating the unstructured: documents go in, filled templates come out, and a human stays in the loop.

RoleFull stack & AI — product, UI, Gemini pipeline

StackNext.js · Python · Gemini

FocusOCR, workflow automation

StatusMVP, in use

Goril Docs dashboard: template management and quick actions

Dashboard: templates, recent documents, quick actions.

The challengeHours of retyping

Operations teams spend hours moving data from PDFs, scans and paper forms into their own documents. It is slow and error-prone, and classic OCR tools either need rigid templates or return a wall of text that still has to be sorted by hand.

The solutionRead, understand, map

Goril Docs uses Google’s Gemini models to read a document the way a person would: it understands the structure, then maps what it finds into fields the user has defined. The output is a filled template, exportable as PDF, DOCX or plain text.

Gemini-powered OCR. PDF, JPG and audio inputs. Falls back between Gemini 1.5 Pro and Flash for accuracy versus speed.
AI field mapping. Extracted text lands in template fields automatically. Two modes: Verbatim copies exactly; Polish lets the model fix grammar and formatting.
Visual layout designer. Upload a PDF background, draw zones on it, and override coordinates by hand to “fix” the AI output visually.

Upload step: a scanned invoice selected, ready for OCR — Upload. Drop a scan, a PDF or a recording.

OCR result: the transcribed invoice text — Transcription. Plain text first, so the user can see what the model saw.

Mapping overview: selected template next to the OCR snapshot — Before mapping: the template and the transcription side by side.

Mapped fields with confidence scores and evidence — After mapping: each field with a confidence score and the line it came from.

Layout designer: zones drawn over a PDF background — Layout designer. Draw zones on the original form; the export renders values into them.

Template library, each with its blank form as background.

The filled template, ready to export as PDF — The result: a filled form, exportable as PDF, DOCX or TXT.

What I learnedThe model is the easy part

Getting Gemini to read a document well took an afternoon. Getting people to trust the result took the rest of the project: showing the source next to the output, making every field editable, and letting the user correct the layout with their hands instead of with a prompt. The human-in-the-loop screens are where the product actually lives.

NextClinico^® →