provenance

Add provenance section below video player in theater mode

final

nnn

readstream

runnies

Apply progressive jitter to pipeline and update existing demo chats

khanmigo

final

origins

re-align

Input a YouTube link and upload a reference document (e.g. policy paper, textbook, research article). The AI will download the video transcript, map it to key themes from the reference document, and generate a synchronized chat replay representing varied internet users (skeptics, hype, experts, etc.).

Simulate realistic audience livestream chat replays based on reference documents

update 3rd sample

Fix YouTube API initialization race condition on simulation switch

Fix Sagan video loading: robust fallback video ID extraction from filename/transcript, SRT file upload support, and privacy-enhanced YouTube host

Update app.py

Add Carl Sagan Demon-Haunted World simulation to choices list and include demo_chat_2.json

Fix tab hover whiteout styling, stop tracking update_demo.py, and rename NtRf4icqE7o srt file

app update

agent update

agent update

frontend update

frontend update

Add custom CSS to resolve whiteout effect when hovering on tabs

Remove invalid theme parameters input_text_color and input_text_color_dark

Configure both light and dark mode variables to force dark theme contrast across all user appearances

Track PDFs using Git LFS and add sample PDFs to the repository

Apply custom theme in gr.Blocks constructor for Hugging Face Spaces compatibility

Fix prompt placeholder to use numeric timestamps and regenerate demo_chat.json

Improve color contrast of badge-mod and usernames to meet WCAG AA guidelines

Regenerate demo_chat.json using updated pipeline and include sample text resources

collapsed stages 1b, 1c, and 2 into a single cohesive generation pass in Stage 2, which runs all segments with global context. I also updated the Stage 3 stylization to perform as a final flourish pass to preserve diversity and usernames

Switch FLASH_MODEL to openai/gpt-oss-120b:fastest

Remove arbitrary document truncation and increase API request timeouts to 300 seconds

Add simulated livestream replay theater, custom generator configuration, and multi-stage LLM pipeline

Add application file

initial commit