AS-IS Document
Generator

The Problem

When onboarding new enterprise clients, HighRadius's implementation engineers had to manually document the client's existing ("as-is") financial processes and Excel workbook usage. This involved sitting through hour-long screen recordings, capturing step-by-step clicks, mapping out data dependencies, and drafting massive Word manuals.

This manual process was incredibly tedious, took several days per client, and delayed overall implementation schedules. It was also prone to omissions, causing configuration errors down the line.

The Solution

I architected and built an automated, multi-modal AI documentation pipeline. Written in Python, the pipeline ingests raw MP4 screen recordings alongside raw Excel workbook metadata. It runs a multi-threaded processing flow to extract frames and sync them with file operations.

Using the Gemini API, the pipeline feeds video frames and workbook metadata to a frame-by-frame LLM sequence. The model acts as an expert analyst: identifying what cells are clicked, which formulas are written, and what financial tasks are being performed.

Finally, a custom Word automation module gathers this structured JSON sequence and compiles it into a professionally formatted Word document containing step-by-step screenshots, cell formulas, and prose explanations, fully formatted in the company style.