Build a Secure Live AI Research Data Processing Prototype

Rapid Low-Code AI Document-Parsing Web Prototype (Secure Cloud Tenant) in Microsoft Copilot Studio / Azure OpenAI

Project Overview:
I need a highly concise, working prototype of a secure web application interface. The sole purpose of this app is to demonstrate a secure, live text-processing pipeline during a brief stakeholder review. It must ingest a raw text document/dataset, simulate a structured evaluation process against a pre-set compliance rubric, and dynamically display an optimization analytics page.

Key Requirements to Implement:

1. Secure File Upload Interface

A minimal, clean user interface with a file drop zone labeled: "Secure Sandbox: Upload Manuscript / Dataset Draft (PDF, DOCX, or CSV)".

A secondary text field labeled: "Alternative Manual Input: Paste Methodology / Text Details".

A dropdown parameter selector: [Parameter Group A / Baseline Metrics Checklist].

2. Localized Governance Engine Setup

Connect this frontend interface securely to an LLM endpoint (via Microsoft Copilot Studio, Azure OpenAI Studio template, or a standard private API pipeline).

Define system prompts to parse the document's sections (checking for structural clarity, data mapping consistency, and alignment with baseline formatting rules).

3. Dynamic Output & Accountability Dashboard
Upon document submission, the UI must refresh dynamically to show:

Maturity / Verification Score: A clean progress gauge (e.g., Ecosystem Readiness Level: 84%).

Compliance Rules Checklist: Visual indicators displaying:

[Passed] Citation & Data Anonymization Check

[Passed] Methodological Parameter Scope

[Warning] Formatting Ambiguity Detected in Section 3

Actionable Next Steps: A generated text box detailing exactly what parameters need optimization.

Technical Constraints:

Must be deployed on a functional, live preview web URL so I can interact with it live during the meeting.

Data privacy configuration must be explicitly set to ensure no data processed is utilized for public model training datasets.

Important Fail-Safe: Hardcode a single specific fallback template path into the interface so that if the live internet connection fluctuates during the review, the target upload file will immediately trigger the exact correct completed dashboard layout cleanly.