Videos ZRM_TfEZcIo

Turn 10,994 Notes Into Memory - Paul Iusztin, Decoding AI & Louis-François Bouchard, Towards AI

index_state ready data_status ok

AI Engineer· published 2026-06-26· 0:39:32· en-US· indexed 2026-08-10 19:54

Open on YouTube

159 shot(s).

keyframes kept every frame deduplicated

What was stored

cues: 362; whisperx 362
chunks: 71; from 362 cues
keyframes: 108; kept of 159 captured
frames with text: 104; 2,809 lines read
chapters: 21; from the source metadata
keyframe bytes: 16.4 MB; word timings on 362 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage	state	model	started	took
`fetch`	done	—	2026-08-10 17:57	1m 48s
`stt`	done	—	2026-08-10 17:58	42s
`chunk`	done	—	2026-08-10 17:59	0s
`text_embed`	done	—	2026-08-10 19:54	1s
`keyframe`	done	—	2026-08-10 17:59	2m 31s
`ocr`	done	—	2026-08-10 18:02	1m 01s
`frame_embed`	done	—	2026-08-10 19:54	19s

Frames, and what the machine read

0:03 #0 done2 line(s)

shot 0·sharpness 2060.4
1. I spent 18 months turning my Second1.00
2. Brain into my living research memory1.00
0:14 #1 done5 line(s)

shot 1·sharpness 2350.5
1. My Second Brain1.00
2. 5,489 notes in Obsidian1.00
3. 5,505 files in Readwise1.00
4. Plus, Notion, Google Drive.1.00
5. Growing ~250 files per month1.00
0:20 #2 done3 line(s)

shot 2·sharpness 1451.8
1. 2026-05-21-pi-repl-vs-tui0.98
2. 2026-05-21-pi-tool-patten0.98
3. Isp-integration0.99
0:25 #3 done1 line(s)

shot 3·sharpness 600.9
1. LaViIS0.54
0:34 #4 skipped

shot 4·duplicate of #2
0:39 #5 done1 line(s)

shot 5·sharpness 831.8
1. Codex, Claude, NotebookLM...0.99
0:43 #6 empty

shot 6·sharpness 552.8
0:51 #7 skipped

shot 7·duplicate of #5
1:05 #8 skipped

shot 8·duplicate of #6
1:13 #9 skipped

shot 9·duplicate of #5
1:30 #10 done1 line(s)

shot 10·sharpness 570.9
1. LOVTES0.72
1:34 #11 done5 line(s)

shot 11·sharpness 3053.5
1. Building Your Own0.99
2. AI Research OS0.96
3. How to turn your Second Brain into a living research memory your agents maintain1.00
4. https://github.com/iusztinpaul/ai-research-os-workshop1.00
5. Paul lusztin· Louis-François Bouchard0.98
1:40 #12 done1 line(s)

shot 12·sharpness 549.5
1. LaVvIE0.53
1:50 #13 done17 line(s)

shot 13·sharpness 2183.6
1. Hi, I'm Paul1.00
2. EXPERT INSIGHT1.00
3. In color1.00
4. LLMEngineer's1.00
5. Handbook1.00
6. Founder & CEO @ Decoding Al0.99
7. Content and courses on shipping Al products1.00
8. Co-Author of the1.00
9. LLM Engineer's Handbook Bestseller1.00
10. Master the art of engineering large1.00
11. language models from concept to production0.99
12. Julien Choumond0.99
13. Co-founder ond CTO, Hugging Face0.87
14. Hamza Tahir0.99
15. Co-founder and CTO, ZEenML0.89
16. Paul lusztin | Maxime Labonne1.00
17. <packt>1.00
2:11 #14 done7 line(s)

shot 14·sharpness 2228.7
1. Hi, I'm Louis-François1.00
2. - Co-founder & CTO@Towards AI0.96
3. — "What's Al” on YouTube · author of Building LLMs for Production0.97
4. - Previously PhD @Mila0.96
5. – Build courses, videos, and trainings for a living0.98
6. OWARDS1.00
7. Which all starts fromgood research!1.00
2:18 #15 done1 line(s)

shot 15·sharpness 138.2
1. TOWARDS NI0.93
2:26 #16 done7 line(s)

shot 16·sharpness 2208.5
1. Hi, I'm Louis-François1.00
2. - Co-founder & CTO @ Towards AI0.94
3. — “What's Al” on YouTube · author of Building LLMs for Production0.97
4. - Previously PhD @Mila0.96
5. – Build courses, videos, and trainings for a living0.99
6. TOWARDSN1.00
7. Which all starts fromgood research!1.00
2:34 #17 done1 line(s)

shot 17·sharpness 137.8
1. TOWARDS AI0.92
2:52 #18 skipped

shot 18·duplicate of #16
3:04 #19 skipped

shot 19·duplicate of #17
3:29 #20 done7 line(s)

shot 20·sharpness 2208.7
1. Hi, I'm Louis-François1.00
2. - Co-founder & CTO @ Towards AI0.94
3. — “What's Al” on YouTube · author of Building LLMs for Production0.97
4. - Previously PhD @ Mila0.94
5. – Build courses, videos, and trainings for a living0.99
6. TOWARDSAI0.98
7. Which all starts fromgood research!0.99
3:57 #21 done7 line(s)

shot 21·sharpness 1571.7
1. Which tool, when?1.00
2. Do you just need a fast answer?0.99
3. – a few quick questions0.98
4. – small source set, no reuse needed0.99
5. – speed matters more than structure0.99
6. → ChatGPT / Claude0.95
7. TOWARDSAI0.96
4:08 #22 done1 line(s)

shot 22·sharpness 199.1
1. TOWARDS AI0.96
4:20 #23 done8 line(s)

shot 23·sharpness 2025.3
1. Which tool, when?1.00
2. Do you just need a fast answer?1.00
3. – a few quick questions0.97
4. – small source set, no reuse needed0.98
5. – speed matters more than structure0.99
6. →ChatGPT / Claude0.98
7. TOWARDSN1.00
8. but — not optimal for longer context or code in general0.99

Transcript

362 cues· 6,459 words· 34,741 chars

chunk 0 · 0:00–0:43 · 135 words · 716 chars
0:00 I spent 18 months turning my second brain into my living research memory.
0:04 Let me explain.
0:05 So within my second brain, I currently have over 5,000 notes in Obsidian and another 5,000 notes in Readwise and some scattered in Notion and Google Drive.
0:15 And all of this is growing on average with 250 files per month.
0:19 And this is what I want.
0:20 On the left, you can see my whole Obsidian vault, this huge mess.
0:24 And whenever I start working on something such as an article, a new project, a new code base, a new feature or whatever, I want to actually pull high signal notes that are actually useful for my current work.
chunk 1 · 0:37–1:18 · 127 words · 665 chars
0:37 And you would ask yourself, why not use directly Codex, Cloud, or Notebook LM?
0:42 And the thing is that I am.
0:43 But you need a system that sits between those harnesses and your second brain.
0:49 Okay, so let's go back to the root of my problem, which is that I'm always losing my research.
0:54 For example, my reading list is a graveyard.
0:57 When I'm scrolling social media and I save that cool X post, a new article, a new YouTube video, a GitHub repository, it doesn't matter.
1:05 Whenever I actually want to start working on something,
chunk 2 · 1:08–1:51 · 133 words · 668 chars
1:08 I never recall what I have in my second brain or
1:11 I have to spend a ton of time actually finding meaningful notes that I can use in my work, right?
1:18 And another problem that I have is that I want the system to actually be anchored into my personal notes, into my personal values, into my personal face.
1:28 I want the system to be personal, to reflect my own thoughts, right?
1:32 And that's why in today's video, Luis Francois and I will teach you how to build your own AI research OS.
chunk 3 · 1:38–2:21 · 129 words · 685 chars
1:38 This also comes with code, so you can also try it out
1:42 yourself.
1:43 And I'm Paul Justine.
1:45 I'm the founder and CEO of Decoding AI, where I do a ton of content on courses on how to ship AI products.
1:51 And I'm also the co-author of the LM Engineers Handbook bestseller.
1:55 And the system, the AI Research OS that I will teach you in this video is the system that I use in my daily work.
2:01 And now I will pass the torch to Louise Francois.
2:05 Thanks, Paul.
2:06 So I'm Louis-François Bouchard.
2:08 I'm the co-founder and CTO of Towards AI, where we build educational courses.
chunk 4 · 2:12–2:46 · 94 words · 499 chars
2:12 And I'm also the creator of What's AI, a YouTube channel where I explain AI engineering techniques I used to explain AI research before.
2:22 now focusing on AI engineering.
2:24 I'm also the author of the book Building NLMs for Production.
2:28 And before that, I was a PhD student.
2:31 So I honestly make research for a living.
2:34 I used to do a PhD, as I said, in AI and doing tons of research and research work.
2:40 Now I build courses, I write videos, I research for videos, I build trainings for companies for a living.
chunk 5 · 2:46–3:30 · 132 words · 672 chars
2:46 And all of these things that I do start with a very good research and also leveraging tons of knowledge and insights that we get at Towards AI from building for clients.
3:01 So I have tons of notes as well, just like Paul, and we try to leverage them the best possible.
3:07 And as you'll see, we built some sort of tool to leverage our second brain where, as you'll see, there will be some differences between how I use it and how Paul uses it.
chunk 6 · 3:17–3:57 · 111 words · 590 chars
3:17 And that's the core goal of the repository that we built on this project is that we want you to adapt it for your needs.
3:24 The whole goal is how can we make research better, but more specifically, how can we better leverage what we have?
3:31 So let's dive into it.
3:33 And first, we need to figure out which tool to use and when, because this whole research system that we built is not for every query.
3:43 If you just need a fast answer, like a few quick questions or just
chunk 7 · 3:49–4:31 · 118 words · 645 chars
3:49 something that you would just Google, basically.
3:52 Well, obviously, just Google it or ask ChatGPT, Cloud, whichever system you want.
3:57 But the problem when doing that is that if you have a lot of following up questions or it's a bigger project that you need to build on
4:06 and have basically a very long context or tons of information to share, relying on ChatGPT isn't ideal.
4:12 And it also means that you are fully dependent on the architecture that OpenAI or ChatGPT's team built.
chunk 8 · 4:20–5:00 · 120 words · 617 chars
4:20 So the next step here is to ask yourself, for a more complex problem, do you need to act quickly or do you want to build some next feature and do something very difficult?

Chapters

0:00 Introduction to the "Second Brain" concept
0:49 The core problem: Losing research and finding meaningful notes
1:32 Building an AI Research OS
2:05 Meet the presenters: Louis-François Bouchard and Pauline
3:31 Choosing the right tools for research (Google vs. LLMs)
4:51 Why NotebookLM and vector databases aren't always ideal
7:08 The need for a personalized research assistant
9:16 Moving data to Obsidian for local file management
11:02 Overview of the AI Research OS repository
12:48 The three-layer system: Raw content, Index, and Wiki
13:28 Architecture of the Deep Research algorithm
18:28 Version 3: Adding the Wiki layer on top of knowledge bases
20:13 How the file-based index works (no database required)
21:26 Exploring the Wiki structure: Comparisons, concepts, and entities
22:49 How to query the Wiki efficiently
25:01 Managing snapshots and personal notes using the PARA method
27:04 Demo 1: Researching agent engineering
31:58 Demo 2: Ingesting and comparing GitHub repositories
34:25 Demo 3: Ingesting custom web links
36:50 Future improvements: Connectors, memory compaction, and source provenance
38:45 The Agent Engineering course overview