Dense Slides: Running AI Application in Minutes w/ AI Templates: Gabriela de Queiroz, Pamela Fox, Harald Kirschner
Source Video
Running AI Application in Minutes w/ AI Templates: Gabriela de Queiroz, Pamela Fox, Harald Kirschner
Method
This deck is slide-only. The existing captured video frame set supplies candidate frames, then local OpenCV rejects sponsor/title/speaker-only frames, crops visible slide surfaces, deduplicates, and saves the cropped slide images.
Cropped Visible Slides
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.99 - Text source: agent_vision.
Slide text:
Microsoft for Startups Founders Hub
Build your startup faster with free access to:
→ Up to $150k of Azure credits and discounts on 30+ tools your team needs to build, like GitHub, Microsoft 365, LinkedIn Premium, and more.
→ Credits toward Azure AI Studio—the most comprehensive set of AI models, including OpenAI GPT-3.5 Turbo, GPT-4, and Llama 2 by Meta.
→ 1:1 sessions with Microsoft experts to help you work through technical and business challenges.
Apply in minutes at https://aka.ms/apply-fh
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.92 - Text source: none.
- OCR decision: ready — UI screenshot with multiple small text regions; OCR will be more reliable than manual transcription in this pass.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.94 - Text source: none.
- OCR decision: ready — Dense UI screenshot with small labels and body text; OCR is the better triage path.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: none.
- OCR decision: ready — Document screenshot with small dense instructions and form-like content; OCR is appropriate.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.95 - Text source: none.
- OCR decision: ready — Portal screenshot with table headers and small interface labels; OCR will capture details better than manual triage.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: none.
- OCR decision: ready — Instruction page with small text, key fields, and endpoint details; OCR is the right triage output.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.94 - Text source: none.
- OCR decision: ready — Application UI screenshot with many small labels and controls; OCR is appropriate for downstream extraction.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.93 - Text source: none.
- OCR decision: ready — Continuation of the playground UI with small interface text; OCR will be more reliable than direct transcription.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.95 - Text source: agent_vision.
- OCR decision: ready — Product UI screenshot with multiple small text regions and interface labels; OCR will capture details better than manual transcription.
Slide text:
Azure AI Proxy Playground
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.95 - Text source: agent_vision.
- OCR decision: ready — Product UI screenshot with dense controls, dropdowns, and response text suitable for OCR.
Slide text:
Azure AI Proxy Playground
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.94 - Text source: agent_vision.
- OCR decision: ready — Same dense interface screenshot style as adjacent frame; OCR is the right extraction path.
Slide text:
Azure AI Proxy Playground
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Documentation page screenshot with a large title and dense table content; OCR is needed for the table and side navigation.
Slide text:
GPT-4 and GPT-4 Turbo models
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: none.
- OCR decision: ready — Continuation of the documentation table with small rows and columns; OCR is appropriate.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Text-heavy registration and SDK instructions with small body copy; OCR will be more reliable than manual transcription.
Slide text:
Registration Details
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Repository README screenshot with title, table, and side metadata; OCR is appropriate for extracting the page structure.
Slide text:
AI App Templates Workshop
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: agent_vision.
- OCR decision: ready — GitHub markdown page screenshot with headings and code/setup instructions; OCR is suitable for the fine text.
Slide text:
Opening the project
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.95 - Text source: agent_vision.
- OCR decision: ready — Product UI screenshot with multiple small text regions and a dense page layout.
Slide text:
Resume codespace
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
- OCR decision: ready — Dense terms page with small body text, form labels, and multiple section headers.
Slide text:
Terms of use
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: none.
- OCR decision: ready — Code editor screenshot with small text, comments, and environment variables that are better handled by OCR.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: none.
- OCR decision: ready — Documentation page with numbered setup steps and command blocks; OCR is the cheapest accurate path.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: agent_vision.
- OCR decision: ready — Dense documentation screenshot with commands and setup instructions; OCR will recover the small text more reliably.
Slide text:
Deploying to Azure
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: none.
- OCR decision: ready — Code editor plus terminal logs with small syntax and runtime output; OCR is appropriate for this dense screen capture.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: none.
- OCR decision: ready — Continuation of the Azure deployment instructions with multiple command snippets and small text.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.99 - Text source: agent_vision.
- OCR decision: ready — Product UI screenshot with dense labels, controls, and chat content that OCR can recover efficiently.
Slide text:
Azure AI Proxy Playground
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: advanced OCR
rapidocr-live/bright-screen/opencv-adaptivereconciled by agent. - OCR decision: ready — Dense table and small UI text; OCR is likely better than direct transcription.
Slide text:
AI App Templates Workshop
These guides will help you get started using our AI App Templates along with the Azure OpenAI Proxy.
Here are the available templates with links to their guides:
Template | Description
OpenAI Chat App | A simple chat app that uses the OpenAI API to generate responses to user input.
RAG on PostgreSQL | A chat app that answers questions using a RAG flow with OpenAI and a PostgreSQL database.
RAG with Azure AI Search | A chat app that answers questions using a RAG flow with OpenAI and Azure AI Search.
We recommend going through each template in the order listed, as they build in complexity.
Workshop instructions: https://aka.ms/ai-workshop
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.95 - Text source: advanced OCR
rapidocr-live/bright-screen/opencv-adaptivereconciled by agent. - OCR decision: ready — Dense UI screenshot with multiple small text regions and code-like output.
Slide text:
RAG on database
OpenAI + PostgreSQL
plz tell me about the best shoes for hiking now
For the best shoes for hiking, we have a few options available:
1. The Trail Xtreme Hiking Shoes by Raptor Elite are a great choice. These shoes are built to endure any trail with their durable leather upper and rugged rubber sole. They offer excellent traction and protection. They come in a timeless brown color that adds a touch of style to your outdoor adventures. The price for these shoes is $135.99.
2. The TrailTracker Hiking Shoes by AirStrider are another excellent option. These shoes are lightweight and breathable, perfect for long hikes. They feature a rugged rubber sole for excellent traction and stability. The cool grey color with green accents adds a touch of style to your hiking ensemble. The price for these shoes is $89.99.
3. If you're looking for something with a steel-blue color, the Trailblaze Steel-Blue Hiking Shoes by WildRunner might be the right choice for you. These shoes are rugged and durable, with a waterproof membrane and a high-traction sole. They are made for superior grip on any terrain. The breathable upper keeps your feet cool and comfortable, and the reinforced toe cap adds extra protection. These shoes are priced at $129.99.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
- OCR decision: ready — GitHub repository page with table entries and sidebar metadata is OCR-suitable.
Slide text:
AI App Templates Workshop
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Blog index page with multiple article titles is OCR-suitable.
Slide text:
Blog Posts
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: agent_vision.
- OCR decision: ready — Dense article text and code block are better read by OCR.
Slide text:
Hybrid Splitters
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.95 - Text source: agent_vision.
- OCR decision: ready — Zoomed article text and code block are OCR-suitable.
Slide text:
Hybrid Splitters
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Chat UI, citation pane, and embedded source page are OCR-suitable.
Slide text:
GPT + Enterprise data | Sample
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.94 - Text source: agent_vision.
- OCR decision: ready — Profile page and repository list are OCR-suitable screenshot text.
Slide text:
Cloud Advocate in Python @ Microsoft
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
- OCR decision: ready — Dense webpage-style screenshot with small body text and an embedded dashboard image.
Slide text:
Monitoring with Application Insights
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
- OCR decision: ready — Code/editor screenshot with a results table and dense small text that is better handled by OCR.
Slide text:
Evaluating a RAG Chat App
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Webpage screenshot with chart and small explanatory text; OCR is more appropriate than manual transcription.
Slide text:
Evaluation
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.99 - Text source: agent_vision.
- OCR decision: ready — Article screenshot with chart, legend, and paragraph text; OCR is suitable.
Slide text:
Hybrid retrieval with semantic ranking outperforms vector-only search
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.99 - Text source: agent_vision.
- OCR decision: ready — Dense article/table screenshot with metric rows and paragraph text that is better suited to OCR.
Slide text:
GPT-based metrics and code-based metrics against RAG chat apps.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: none.
- OCR decision: ready — Cropped dense table and paragraph text; OCR is needed to recover the full content.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: none.
- OCR decision: ready — Dense article screenshot with text and bullet content; OCR is needed for reliable extraction.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Webpage screenshot with multiple text regions and sidebar content; OCR is more appropriate than manual transcription.
Slide text:
Welcome to Apache Lucene
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Dense browser/search-results screenshot with small text and multiple panels.
Slide text:
Okapi BM25 - Wikipedia
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
- OCR decision: ready — Dense article screenshot with a paragraph and table; OCR is better than manual transcription here.
Slide text:
Metric computed with vector retrieval only (no Semantic ranking).
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.93 - Text source: none.
- OCR decision: ready — Dense multi-pane demo/code screenshot with small labels, terminal output, and editor text.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.94 - Text source: none.
- OCR decision: ready — Dense IDE and chat screenshot with code, tabs, and small UI labels.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.94 - Text source: none.
- OCR decision: ready — Dense IDE and chat screenshot with explanatory text and code on screen.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.95 - Text source: none.
- OCR decision: ready — Dense benchmark/documentation screenshot with small text and code-like markup.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.99 - Text source: agent_vision.
- OCR decision: ready — Dense package page screenshot with sidebar navigation, descriptive text, and links.
Slide text:
Azure AI Inference client library for Python
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.99 - Text source: agent_vision.
- OCR decision: ready — Dense package page screenshot with sidebar navigation, descriptive text, and links.
Slide text:
Azure AI Inference client library for Python
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
- OCR decision: ready — Dense product/UI screenshot with many small text regions; OCR will be better than manual transcription.
Slide text:
Azure AI Inference client library for Python
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: none.
- OCR decision: ready — Dense terminal/CLI help output with small text; OCR is appropriate.
Classification audit: raw/sources/slide-ai-classification/dense/1t-9-s1brcg/audit.json