DocuWrangler: AI Wrangler of Annual Reports

Fetches insights from dense corporate documents through an immersive, music-enhanced VR experience.

Inspiration

Annual reports are dense, jargon-heavy, and often overlooked despite containing critical financial and strategic insights. We set out to reimagine how stakeholders interact with these documents by turning the process into an immersive Wild West–themed experience. DocuWrangler was built to make reading and understanding annual reports both engaging and accessible.

What it Does

Enables natural language questions about uploaded annual reports
Returns accurate, context-aware answers with specific page references
Supports content extraction from text, tables, and data visualizations
Allows users to enter a Unity-based VR saloon environment
Classifies response tone and triggers genre-matching AI-generated music

How We Built It

PyMuPDF to extract and preprocess data from the annual report PDF
LangChain and GPT-4o to build a retrieval-augmented Q&A system using ChromaDB
Flask backend to serve responses with tone and page metadata
SOUNDRAW used to pre-generate and categorize music tracks
Unity AudioSource to handle in-VR audio playback based on tone
Unity VR for the front-end saloon interface

Challenges We Ran Into

Extracting structured data like tables and charts from unstructured PDF formats
Preventing AI hallucinations and ensuring factual grounding to report content
Citing the source for the generated answer and the confidence score.
Designing an immersive user experience while maintaining business relevance

Accomplishments That We're Proud Of

Developed a working end-to-end VR system that bridges generative AI and music with enterprise reports
Successfully mapped PDF content to accurate answers and visual/audio cues
Handled multiple data types: raw text, tables, and image-based figures
Delivered a unique storytelling layer to corporate document analysis