Product Design and AI Development: Form-to-Document AI Orchestration

How I delivered a modular AI pipeline using Power Automate and OpenAI to compile structured inputs into user-ready documents, with robust error handling and formatting control.

By Joseph Arnold

I delivered a modular AI pipeline using Power Automate, Microsoft Forms, and OpenAI’s API to handle rate limits, token constraints, and multi-part prompts, which then dynamically compiled results into user-ready documents. I built robust error handling and formatting control logic for reliability.

Core Problem

1

I Identified a Lack of Multi-Input UX

I saw that end users needed to submit structured questions and documents across categories, but existing chat interfaces lacked the multi-input UX and formatting control required for complex prompts.

2

I Solved for Missing Guided Workflows

Before document upload was mainstream, I recognized that users had no system to consolidate inputs and generate rich, actionable outputs in a repeatable way, so I built one.

3

I Designed a Unified, No-Code Interface

I determined that a unified, no-code interface was needed to collect structured information, guide user expectations, and return high-quality outputs tailored to different task types.

Strategic Objectives

Build Accessible Interface
  • I built an accessible, extensible prompt delivery interface using Microsoft Forms as a structured front end.
  • I enabled users to request one of several document-generation or advisory modes (e.g. writing assistant, decision support).
  • I returned formatted, branded, and modular document outputs, which I automated end-to-end via Power Automate and the OpenAI API.

Key Initiatives

Microsoft Form UX and Intake Logic
  • I designed a multi-mode input form to support use cases like writing assistance, decision analysis, or quiz-style intake.
  • I included input sanitization and structured fields (e.g. article drafts, decision context, constraints, goals, tone preferences).
OpenAI Prompt Flow and Token Control
  • I built a modular prompt architecture for each mode, including for content editing and values-aligned decision-making.
  • I implemented token-aware segmenting logic to ensure prompt content remained under limits and output remained cleanly structured.
  • I handled multi-query orchestration and response merging to ensure a seamless user experience.
Backend Orchestration (Power Automate)
  • I authenticated and sent POST requests to OpenAI’s API using custom JSON payloads and handled multi-part input variables.
  • I sanitized input characters to prevent encoding errors or markdown corruption in responses.
  • I designed the system to parse and recompile model output into a branded Microsoft Word template.
Branded Document Output and Delivery
  • My system applied consistent style and structure across generated PDFs regardless of content type.
  • I automated the conversion to PDF and delivery via a customized thank-you email.
  • I built support for diverse content types, including editing advice, Q&A packets, and personalized coaching briefs.

Executive Summary

For this initiative, I created a flexible, low-code blueprint for advisory automation and multi-modal document generation. By partnering directly with developers to frame ambiguous friction as user stories with measurable ROI, I reduced friction for users with complex or multi-variable input needs. My solution enabled non-technical testers to engage with AI models in a structured, safe, and consistent format without manual copy/paste or formatting between tools.

  • I enabled a single-form UX for high-trust document output.
  • I secured engineering commitment by tying the pilots I designed to operational efficiency outcomes.
  • I designed the system for enterprise extensibility with a unified delivery layer and modular prompt adaptation.