Dense Slides: From 46% to 90%: Fine-Tuning Tiny LLMs for On-Device Agents — Cormac Brick, Google

Source Video

From 46% to 90%: Fine-Tuning Tiny LLMs for On-Device Agents — Cormac Brick, Google

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_slide confidence 0.98
Text source: advanced OCR rapidocr-live/border-trim/contrast reconciled by agent.
OCR decision: ready — Dense small text in four cards; OCR is likely more accurate than manual triage.

Slide text:

Running AI on the Edge has many benefits

Latency / UX

Faster, no network involved

Privacy

Sensitive data not sent off device

Offline use

Works without requiring cellular

Savings

Lower or no data-center costs

AI Engineer EUROPE

Classification audit: raw/sources/slide-ai-classification/dense/-TiET_K-E_g/audit.json