---
title: "Reconstructed Slides: From 46% to 90%: Fine-Tuning Tiny LLMs for On-Device Agents — Cormac Brick, Google"
category: "slides"
video_id: "-TiET_K-E_g"
sourceLabels: ["Cropped public YouTube video frames", "Local OpenCV slide-region detection", "Local RapidOCR"]
---

# Reconstructed 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](https://www.youtube.com/watch?v=-TiET_K-E_g)

## Method
This deck is reconstructed from the existing video frame captures by detecting likely slide regions with OpenCV, cropping/upscaling those regions, deduplicating similar crops, and OCRing the cropped slide images locally. It is a cleaner companion to the full-stage frame deck.

## Reconstructed Slides
![[assets/reconstructed-slides/-TiET_K-E_g/slide-001.jpg]]

- Source frame: `slide-001.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `157.79`
![[assets/reconstructed-slides/-TiET_K-E_g/slide-002.jpg]]

- Source frame: `slide-002.jpg`
- Crop: `contour` `[0, 0, 960, 540]` score `161.12`
![[assets/reconstructed-slides/-TiET_K-E_g/slide-003.jpg]]

- Source frame: `slide-003.jpg`
- Crop: `contour` `[0, 0, 960, 540]` score `176.1`
![[assets/reconstructed-slides/-TiET_K-E_g/slide-004.jpg]]

- Source frame: `slide-004.jpg`
- Crop: `contour` `[0, 0, 960, 540]` score `166.32`
![[assets/reconstructed-slides/-TiET_K-E_g/slide-005.jpg]]

- Source frame: `slide-005.jpg`
- Crop: `contour` `[0, 0, 960, 540]` score `166.43`
![[assets/reconstructed-slides/-TiET_K-E_g/slide-006.jpg]]

- Source frame: `slide-006.jpg`
- Crop: `contour` `[0, 0, 960, 540]` score `167.04`
![[assets/reconstructed-slides/-TiET_K-E_g/slide-007.jpg]]

- Source frame: `slide-007.jpg`
- Crop: `contour` `[0, 0, 960, 540]` score `166.46`
![[assets/reconstructed-slides/-TiET_K-E_g/slide-008.jpg]]

- Source frame: `slide-008.jpg`
- Crop: `contour` `[0, 0, 960, 540]` score `166.77`
![[assets/reconstructed-slides/-TiET_K-E_g/slide-009.jpg]]

- Source frame: `slide-009.jpg`
- Crop: `contour` `[0, 0, 960, 540]` score `176.67`
## Dense Scene-Detected Slide Candidates
- [[youtube--TiET_K-E_g-dense-slides]]