Dense Slides: WF26: Harness Engineering & Startup Battlefield ft. Garry Tan, Mike Krieger, @t3dotgg , DSPy
Source Video
WF26: Harness Engineering & Startup Battlefield ft. Garry Tan, Mike Krieger, @t3dotgg , DSPy
Method
This deck is slide-only. The existing captured video frame set supplies candidate frames, then local OpenCV/RapidOCR rejects sponsor/title/speaker-only frames, crops visible slide surfaces, deduplicates, and OCRs the cropped slide images.
Cropped Visible Slides
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: none.
- OCR decision: ready — Chart labels and callout text are dense.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: none.
- OCR decision: ready — Architecture diagram and small labels are OCR-suitable.
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Stacked diagram labels are OCR-suitable.
Slide text:
The Intelligence Layer
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
Slide text:
RL Selects the Response, Not the Facts
State:
1. Failure category
2. Risk level
3. Retry count
4. Drift severity
5. Data-quality condition
Action:
retry • coerce • rollback • quarantine • escalate • log
1. Tabular Q-learning
2. Small, interpretable state space
3. Low-memory inference
4. Inspectable Q-values for every decision
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
Slide text:
(problems) One conversation has to do everything
Meaning:
the context window becomes
- the database,
- the file system,
- the memory,
- the reasoning space
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: agent_vision.
- OCR decision: ready — Browser screenshot with several small labels and footer text.
Slide text:
Track narrative state, not numbers
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: agent_vision.
- OCR decision: ready — Screenshot text and small captions are OCR-suitable.
Slide text:
Narrated on the fly · rendered to paper
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
- OCR decision: ready — Dense slide with chart and small body text.
Slide text:
Your context window has a hidden RAM bill
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
- OCR decision: ready — Diagram-heavy slide with small explanatory text.
Slide text:
What is TurboQuant
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.99 - Text source: agent_vision.
Slide text:
what people think the agent is
model
runtime / sandbox
tools
loop
framework
- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
- OCR decision: ready — Code slide with multiple lines and small text.
Slide text:
Simplified Loop
Hidden Non-Slide Evidence
- `slide-005.jpg` —
speaker_stageconfidence0.99; speaker intro card
Classification audit: raw/sources/slide-ai-classification/dense/I2cbIws9j10/audit.json