01 / Flagship · In development

FormMind

An in-development AI document automation project for pre-qualification forms, supplier questionnaires and tender responses in Word and Excel—built to reuse trusted knowledge without turning review into a black box.

Document intelligenceHuman reviewKnowledge reuse
Period2025–present
RoleProduct concept, workflow and build
OriginCursor Hackathon Singapore
StatusIn development

The problem

Teams keep answering the same difficult questions inside different files.

Engineering and fabrication teams receive pre-qualification forms, supplier questionnaires and tender responses in inconsistent Word and Excel layouts. Useful answers already exist, but they are scattered across old submissions, supporting documents and individual memory.

Copy-and-paste is slow and hard to verify. A generic chatbot can draft quickly, but it often removes the evidence trail and does not respect the structure of the file that must actually be returned.

The workflow

The intended workflow keeps judgment visible.

  1. 01

    Use the real document

    The intended workflow starts with the Word or Excel file people already use—not a replacement web form.

  2. 02

    Retrieve trusted knowledge

    The system is being designed to find relevant previous answers and supporting knowledge for each requirement.

  3. 03

    Review visibly

    Suggestions are intended to show their source and confidence so a person can accept, edit or reject them.

  4. 04

    Return a usable file

    The target output is an approved file whose original structure and formatting remain protected.

Current build focus

The product is being rebuilt around document fidelity.

01

Verified fixture set

Real Word and Excel examples are used to test whether generated files preserve the source document.

02

Grounded answer approach

Previous answers and company knowledge are being structured as visible evidence rather than invisible prompt material.

03

Human decision layer

The review experience is being designed so suggestions can be accepted, edited or rejected field by field before export.

04

Safe output path

Current engineering work focuses on protecting layout, formulas, formatting and the original source files.

Next projectLifeOS — In progress