← Back to work
Intelligent Form System
Agentic AIMCPLLMsFull-stack

Intelligent Form System

2025Internal tool

An internal AI workflow that turns documents, links, or plain text into structured records, asks for missing information, and executes an approved action after human review.

The problem

Creating records such as training modules required people to collect information from documents, links, or messages and enter it into the platform by hand.

How it works

  1. 01, Input
    Unstructured input

    A document, a link, or pasted text.

  2. 02, Model
    Field extraction

    An LLM pulls out the structured fields the record needs.

  3. 03, Model
    Follow-up questions

    Asks the user for anything that is missing.

  4. 04, Person
    Human review

    The user inspects the result, requests changes, or approves it.

  5. 05, Tool or code
    Approved action

    MCP tools execute the action and write the record to the database.

↺ Changes requested: back to step 02
InputModelPersonTool or codeNothing is executed until a person approves it.

What I did

  • Built the agentic workflow, MCP integrations, and the front-end interface used by the internal team.
  • Extracted structured fields from a document, link, or pasted text and asked follow-up questions when information was missing.
  • Added a review loop so users can inspect the result and request changes before approval.
  • Executed the approved action, including writing the final record to the database.

Results

  • Used by the internal team to create real records in the training platform.
  • Keeps a person in control before any action is executed.
← All projects