# Dhanya808

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Dhanya R M](https://cerebralvalley.ai/u/Dhanya808)
- **GitHub:** https://github.com/dhanya808/deepdoc-studio
- **Demo video:** https://youtu.be/QzD4dN5IJco
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1886

Standard OCR pipelines and conventional LLMs treat complex documents as flat text strings. When given financial 10-K earnings filings, scientific research papers, or clinical trial reports, this text-flattening approach catastrophically scrambles tabular columns, corrupts mathematical equations, and destroys the spatial context connecting visual charts to numerical data.

DeepDoc Studio solves this by treating documents as visual scenes rather than flat text. Powered by Meta SAM 3.1, it executes zero-shot semantic visual segmentation across pages, isolating tables, figures, charts, and equations into atomic visual regions with real-time confidence scores.

These visual regions are dynamically routed to Meta's specialized models:
1. Meta Muse Spark powers the Table Studio (converting dense tables into live, sortable data grids with instant export to CSV, JSON, Pandas, and SQL DDL) and decompiles mathematical formulas into clean LaTeX with variable semantic dictionaries.
2. Meta Muse Wonder enables multi-page cross-modal reasoning (e.g. correlating an infrastructure CapEx chart on Page 2 with balance sheet rows on Page 1), complete with step-by-step reasoning chains and clickable spatial citation badges.
3. Automated Executive Brief generation produces audit-ready summaries with single-click export.

## More from Meta Global AI Developer Hackathon

- [Pawlink Team](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1883)
- [HumanExperiences](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1884)
- [`Book by the People`.](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1885)
- [AI Innovators](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1887)
- [SpriteForge](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1888)
- [Bluecoderhub](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1889)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1886. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt