YouTube2Text

สร้างแอป Local AI แปลงไฟล์ PDF เป็น Excel ภาษาไทยไม่เพี้ยน! — Transcript

by BoomBigNose · 3,484 words · 648 segments · language en · Watch on YouTube

Full transcript

  1. 0:00So, here, OCR only takes about 200
  2. 0:03milliseconds for 38 lines.
  3. 0:07You can save it to SQL or export it to
  4. 0:09Excel.
  5. 0:14Wow, this program is amazing! Hello
  6. 0:16everyone, this is Broom.
  7. 0:19Today I'm showing you how to create a
  8. 0:21Local AI harness for accounting.
  9. 0:26We can receive PDF files, scanned bills
  10. 0:29/receipts, and photos—all within this
  11. 0:31harness using GoLanguage from Google.
  12. 0:39Today, the big picture involves 5 steps
  13. 0:41: receiving files, preparing files,
  14. 0:44reading documents, and this is the
  15. 0:46highlight: the OCR model in
  16. 0:48ThailandCommunity has just been updated
  17. 0:51, and it's a great model, like Typhoon
  18. 0:54OCR or OpenThai 2.0, the hottest OCR
  19. 0:56models right now.
  20. 1:01So, today I'll show you how to install
  21. 1:03it using HuggingFace.
  22. 1:08I'll use one first to test the flow here
  23. 1:10.
  24. 1:12We'll also organize the program's table
  25. 1:14, columns, and check the formatting,
  26. 1:17because when we scan hundreds of files,
  27. 1:19bills, and receipts, we need to organize
  28. 1:21the table.
  29. 1:24Then we can export it to Excel.
  30. 1:27It can be a CSV file or a local SQLite
  31. 1:30file.
  32. 1:31This is another step in the backup
  33. 1:33process.
  34. 1:35Here, we'll deal with data hashing,
  35. 1:38cipher encryption, and chart
  36. 1:39verification of the files.
  37. 1:44This system helps make the harness
  38. 1:52or equipment used by each organization
  39. 1:54and business more secure.
  40. 1:59Okay, I'll open the slides and explain
  41. 2:02step by step what a local AI harness is
  42. 2:04.
  43. 2:10A harness allows us to change any model
  44. 2:12because a harness is like everyone
  45. 2:14building a basic engine structure.
  46. 2:20For example, if I build a car or a robot
  47. 2:23, I can change the engine, the brain
  48. 2:25inside, whether it's Typhoon, OpenThai,
  49. 2:28or various models like Google Gemini,
  50. 2:31GPT, etc.
  51. 2:39I'll show you how to use the antigravity
  52. 2:41model because it's a brand new model,
  53. 2:44Gemini 4.0, which includes carbon and
  54. 2:46hydrogen.
  55. 2:50I'll be using these two together.
  56. 2:53This one, hydrogen, is an orchestator.
  57. 2:59Actually, an orchestator is, I'll
  58. 3:01explain briefly, it's used to structure
  59. 3:04and distribute tasks to AI.
  60. 3:07It's called a sub-agent or agent team.
  61. 3:12It can be used, but I recommend using it
  62. 3:14for large projects, especially if you're
  63. 3:16running it in parallel.
  64. 3:22As I said, carbon and hydrogen really
  65. 3:24work well together.
  66. 3:28And there's a free quota from Google too
  67. 3:30, so you can test it out.
  68. 3:32I'll provide the download link here.
  69. 3:35You can download Google Antigravity
  70. 3:37according to your OS.
  71. 3:41Once finished, we'll get a page like
  72. 3:43this.
  73. 3:45Okay, let's start the steps.
  74. 3:48We'll use Google Antigravity, either the
  75. 3:51Go language or Python.
  76. 3:54Then, we'll follow these 5 steps.
  77. 3:59I'll keep it simple for everyone.
  78. 4:02You can copy the files in this slide or
  79. 4:05like this.
  80. 4:08I'll capture this and use it as a
  81. 4:10project folder for this application.
  82. 4:17Then, click on Google Antigravity, and
  83. 4:19then click the plus sign (the one I use
  84. 4:21when creating any application) and
  85. 4:23select "from this folder."
  86. 4:30On my computer, I'll create it on the
  87. 4:32desktop and name it "bin."
  88. 4:38Okay, so we're using billingPDF, text,
  89. 4:42OCR, and local AI harness.
  90. 4:50It might be a bit long, but we can
  91. 4:51shorten it to this.
  92. 4:55Okay, then I'll click create and open,
  93. 4:58and then click next.
  94. 5:02Now we'll be in our local project folder
  95. 5:04on our desktop.
  96. 5:08After that, this will be the worktree.
  97. 5:11Let me explain it simply.
  98. 5:14It's not complicated at all.
  99. 5:19We already have the main code, and if we
  100. 5:21want to open the worktree without
  101. 5:23affecting the main code, we open a
  102. 5:25branch, do the work, and then merge it
  103. 5:27into the main branch.
  104. 5:31What's the benefit of this?
  105. 5:33It's beneficial...It's very important
  106. 5:36when we run parallel applications,
  107. 5:38creating multiple features
  108. 5:39simultaneously or working on separate
  109. 5:42tasks—meaning running them in parallel—
  110. 5:44using orchestrators or agents.
  111. 5:48This relates to worktrees.
  112. 5:52But today, for the fastest applications
  113. 5:55we can start with, I recommend avoiding
  114. 5:57branching into lines, placing
  115. 5:59orchestrators, and not placing agent
  116. 6:01teams yet.
  117. 6:05I strongly advise against this, based on
  118. 6:07research, as it reduces performance
  119. 6:09unless the task is complex or requires
  120. 6:11running multiple features concurrently.
  121. 6:17Okay, I'll click the plus button to
  122. 6:19upload the file.
  123. 6:22This is my screen sharing right here.
  124. 6:25This is the application file, everyone.
  125. 6:29This one, and after that, we can
  126. 6:31actually
  127. 6:36use it, or we can use it first.
  128. 6:42I'll let you know first whether this
  129. 6:43idea is good or not, and whether it will
  130. 6:45really meet the needs.
  131. 6:51Then, everyone, I'm going to adjust it
  132. 6:53to turbo because it focuses on speed,
  133. 6:56and the data is relatively fine, so we
  134. 6:58can approve it.
  135. 7:04Turbo, okay?
  136. 7:07Let's
  137. 7:10talk
  138. 7:13about
  139. 7:16the
  140. 7:20all-website,
  141. 7:24or the structure.
  142. 7:32I'll tell you whether this structure
  143. 7:34meets the needs of our organization, our
  144. 7:36system.
  145. 7:41For example, we use Windows or Not,
  146. 7:43something
  147. 7:49like that for now.
  148. 7:52Then, I'll use hydrogen first, and use
  149. 7:55it because there are four hydrogens, it
  150. 7:57won't consume as much as carbon.
  151. 8:02Carbon is very smart, but during this
  152. 8:04phase, or asking about our idea, I
  153. 8:06recommend using something that isn't too
  154. 8:08focused on coding, and that you already
  155. 8:11have a reasonable amount of knowledge
  156. 8:13about, it will meet the needs better.
  157. 8:19Then, while they're guild-ing, I'll read
  158. 8:21this first because sometimes they'll get
  159. 8:23a pop-up like this.
  160. 8:33Okay,
  161. 8:38okay.
  162. 8:40And here, actually, if we use it for
  163. 8:42multi-users and share an office PC, I
  164. 8:44would highly recommend this because it's
  165. 8:47for real work, for work used in a real
  166. 8:49organization.
  167. 8:53And, or, this one...For running the
  168. 8:56project, if I were to recommend one, I'd
  169. 8:58suggest option two.
  170. 9:02However, for maximum convenience right
  171. 9:05now, I'd recommend option one.
  172. 9:08Everyone, if you want to delve deeper
  173. 9:11into the full-fledged option two for
  174. 9:13organizational use, you can contact me
  175. 9:15directly here to discuss and ask
  176. 9:17questions.
  177. 9:21You can also consult with me about the
  178. 9:23community
  179. 9:31version, which includes all the source
  180. 9:33code and class materials
  181. 9:41that you can download and use directly.
  182. 9:48A new version will be available soon;
  183. 9:50you can join via the link below.
  184. 9:56If everyone wants to consult me
  185. 9:58initially, they can do so normally.
  186. 10:03Okay, so we'll look at this application
  187. 10:05.
  188. 10:07It has five different system components
  189. 10:09, and I'll explain further.
  190. 10:17As I said, it's like a car or a robot
  191. 10:19with a complete frame that can run just
  192. 10:21by adding an engine and a brain.
  193. 10:28How is this different from other
  194. 10:30software applications?
  195. 10:33It can think, has memory, and can handle
  196. 10:35all kinds of devices.
  197. 10:39For example,
  198. 10:44it
  199. 10:52can create programs for accounting and
  200. 10:55other tasks.
  201. 11:04This is why it's called...well, as I
  202. 11:06explained, it can read files, prepare
  203. 11:08images, and has all the necessary
  204. 11:10equipment for reading bills and
  205. 11:12documents, preparing them for export,
  206. 11:14and connecting models.
  207. 11:27This video or this application can do
  208. 11:29various things: it can take photos,
  209. 11:31convert image files, correct tilting,
  210. 11:33call models, and then check the table
  211. 11:35arrangement to see if the data is
  212. 11:37correct according to the columns and
  213. 11:39columns.
  214. 11:46We check and save the data locally or
  215. 11:48elsewhere so we can use it later.
  216. 11:54It's important to be mindful of our
  217. 11:56personal data and not forget security.
  218. 11:58This is crucial because otherwise, our
  219. 12:00financial information could be leaked.
  220. 12:05Another important reason why we should
  221. 12:07be careful about our financial and
  222. 12:09accounting data is that this information
  223. 12:11can be sent to other servers and
  224. 12:17processed by various companies.
  225. 12:22AI doesn't just think about what a
  226. 12:24company's bill is; it can analyze how
  227. 12:26much a business typically spends on that
  228. 12:28type of expense.
  229. 12:32Our data could be publicly disclosed.
  230. 12:35So, I want everyone to be aware of local
  231. 12:38AI and use it wisely, without wasting
  232. 12:40tokens, and to respect privacy.
  233. 12:43I'll discuss this further in the next
  234. 12:45section. Okay, let's move on. I'll
  235. 12:47explain the importance of local AI with
  236. 12:49privacy.
  237. 12:52We can fine-tune models without paying
  238. 12:55for expensive subscription courses, GPT
  239. 12:57chat, or Gemini.
  240. 13:01We can take those and build a structure
  241. 13:03, then create our own harness and local
  242. 13:05AI for use.
  243. 13:11If you want to test this individually or
  244. 13:13discuss it further, you can join the
  245. 13:15community.
  246. 13:17But if you want my company or my team to
  247. 13:19help set up this system, you can contact
  248. 13:21us.
  249. 13:24I also have harness systems for
  250. 13:26organizations.
  251. 13:31Okay, so regarding the model size, I'll
  252. 13:33elaborate on the model size we can...
  253. 13:36The OCR functionality can be changed
  254. 13:38here.
  255. 13:41In the accounting field, there are two
  256. 13:44latest models: the SCB Typhoon and
  257. 13:46OpenThai 2.0.
  258. 13:49Both are excellent and developed by
  259. 13:51incredibly talented researchers in
  260. 13:53Thailand.
  261. 13:54Don't forget to like, share, and add
  262. 13:56them on Facebook.
  263. 13:59I'll provide the source code and give
  264. 14:01credit to the development team.
  265. 14:050.8 B refers to the model size of this
  266. 14:08AI model from SCB Typhoon/OpenThai. It's
  267. 14:11lightweight and fast.
  268. 14:130.8 B refers to the parameter size; this
  269. 14:16one has 800 million parameters, and this
  270. 14:18one has 2 billion parameters.
  271. 14:23In simple computer terms, this means the
  272. 14:26model size is approximately 1.3 GB
  273. 14:29versus 3.2 GB.
  274. 14:35The model's clarity is 8-bit, and if
  275. 14:37you're using a Mac, I recommend 8GB of
  276. 14:40RAM or more (16GB or more if you're
  277. 14:43using a 2-bit system) and a graphics
  278. 14:46card with 4GB of VRAM or more.
  279. 14:49There's a formula for calculating VRAM,
  280. 14:52it might not be perfectly precise, but
  281. 14:54it helps a lot.
  282. 14:56The safest approach is to have plenty of
  283. 14:59VRAM and plenty of RAM.
  284. 15:03Macs use processors that
  285. 15:09can integrate all their GPU cores and
  286. 15:11RAM.
  287. 15:14PCs without dedicated graphics cards now
  288. 15:16have a similar structure to Macs.
  289. 15:20Newer Spark notebooks are starting to
  290. 15:23meet these needs, so you don't have to
  291. 15:25worry; they'll be similar to Macs.
  292. 15:29Okay, so...There's also a rating system
  293. 15:32for how the Thai language scans perform
  294. 15:34, and this will be ours.
  295. 15:39Let's continue.
  296. 15:41Actually, everyone needs to install it
  297. 15:43first, or you can use Python.
  298. 15:48It's not a problem at all, but it's
  299. 15:50lighter in terms of running speed and
  300. 15:52application size.
  301. 15:56It will run more smoothly.
  302. 15:58Actually, it's very simple.
  303. 16:02Everyone can take my slides, the
  304. 16:04commands below, the slides that are
  305. 16:06already in the application, and paste
  306. 16:08them like this.
  307. 16:12It's like the
  308. 16:15AI is our assistant. Sometimes we
  309. 16:17understand the management of commands,
  310. 16:19understand the structure, and know which
  311. 16:21app to use; we just let it handle the
  312. 16:23rest. Yes, it can be done.
  313. 16:27For UI, I'll use this as a
  314. 16:28recommendation here, but I need to
  315. 16:31mention that right now I'm using macOS.
  316. 16:35Also, please use the "give me version"
  317. 16:38in macOS.
  318. 16:40Actually, you don't need to include it;
  319. 16:42I'll remove it.
  320. 16:46I didn't continue this because I wanted
  321. 16:48to say that we're currently using macOS
  322. 16:50.
  323. 16:52Okay, everyone, don't forget to install
  324. 16:55the Go version. If you want to use it
  325. 16:57like me, copy the command here in the
  326. 16:59slide and paste it.
  327. 17:02The next part, the model runner, is
  328. 17:05quite important: LMStudio or lama.cpp.
  329. 17:10Actually, if everyone is using an NVIDIA
  330. 17:13graphics card or that architecture, you
  331. 17:16can use VLLM.
  332. 17:18Here's VLLM.
  333. 17:22Everyone, it will be...It's about
  334. 17:25running AI models smoothly, fast, easy,
  335. 17:27fast, and cost-efficient for NVIDIA
  336. 17:30architecture.
  337. 17:33For example, CUDA, everyone, CUDA is
  338. 17:36from NVIDIA.
  339. 17:39CPUs also have it, but for AI to really
  340. 17:41run, it's better to run it on CUDA on
  341. 17:44the GPU, also from NVIDIA.
  342. 17:47CUDA, everyone.
  343. 17:49Sorry if I mispronounced it.
  344. 17:51Okay, then load Typhoon OCR, choose the
  345. 17:54model size, or use OpenType.
  346. 17:58For example, I have 48GB of RAM on my
  347. 18:01MacBook, so I'll use a 2 billion model.
  348. 18:07Let me check which one to load first.
  349. 18:11I actually want to load OpenType too.
  350. 18:14Let me look for a moment. Okay, I'll use
  351. 18:17OpenType. This one.
  352. 18:19And then...So, we'll have Olama pull it
  353. 18:22down.
  354. 18:24We need to check if Olama has it or not
  355. 18:26.
  356. 18:27For example, with Typhoon, they've
  357. 18:29deployed it, and it's great! Everyone
  358. 18:31has it in Olama.
  359. 18:33This part is also in Olama, so it's easy
  360. 18:36and convenient.
  361. 18:38You can download it here.
  362. 18:40The command is in the slide as well;
  363. 18:42just copy it from here.
  364. 18:44Everyone, finish the first line, then
  365. 18:46check and type the second line in the
  366. 18:48terminal like this.
  367. 18:51I'll show you the terminal window. It
  368. 18:53will look like this. Everyone, the
  369. 18:55terminal and
  370. 18:58Olama are okay.
  371. 19:01No problem, I'll use this one first
  372. 19:03because it's very easy for Olama.
  373. 19:07Olama isn't installed yet; you can
  374. 19:09install it directly through the download
  375. 19:11here. It's compatible with MacOS, Linux
  376. 19:14, and Windows. Here are the download
  377. 19:17links, everyone.
  378. 19:18Place them here, then I'll install the
  379. 19:21Olama file first.
  380. 19:24Okay, then check here for any issues.
  381. 19:26Follow their instructions, they
  382. 19:28recommend it, and then submit.
  383. 19:33Next, type the Olama Pool OCR model here
  384. 19:36for testing.
  385. 19:40My computer has a lot of RAM, but
  386. 19:42version 0.8 is fine for now, it should
  387. 19:45be easy.
  388. 19:46Everyone can run it too.
  389. 19:48But if you have a powerful computer, use
  390. 19:51this link, everyone. This link is for
  391. 19:54version 2.0.
  392. 19:56I'll check with the Thai open forum
  393. 19:58first to see if it's been deployed yet.
  394. 20:01There might be, everyone seems to know.
  395. 20:06I'll check with the Thai open forum,
  396. 20:08Professor Kae.
  397. 20:12This is really important, especially the
  398. 20:14popular and newest parts. Oh, they're
  399. 20:16not here yet, everyone. That's okay,
  400. 20:19it's not too serious.
  401. 20:21We'll consider this as a use case
  402. 20:23example that everyone can adapt
  403. 20:24themselves.
  404. 20:27You can download it from here or
  405. 20:29directly from [link].
  406. 20:32I'll skip this part for now.
  407. 20:35This
  408. 20:38model is approved.
  409. 20:41Okay, I'll finish step 1 first, then
  410. 20:43I'll go to the model above.
  411. 20:47It's simple, everyone.
  412. 20:51For this application, use
  413. 21:00this model.
  414. 21:02Answer their questions, and that's it.
  415. 21:05Oh, this is still in the process.
  416. 21:09I'll send the model for them to run.
  417. 21:12Wow, everyone, it's amazing!
  418. 21:15Actually, regarding the 4-digit code, I
  419. 21:18use all of them.
  420. 21:25Everyone, both the Google code and the
  421. 21:27model from the company that's always
  422. 21:29popular, but I'm really starting to like
  423. 21:31Google because you get everything from
  424. 21:33Google itself, like YouTube Premium.
  425. 21:41It's a huge ecosystem.
  426. 21:44If you subscribe to Google's code, and
  427. 21:49maybe a little bit from other things,
  428. 21:51maybe get a discount or something.
  429. 21:53This is my plan.
  430. 21:56That's about it.
  431. 21:59I'll look at the next steps. Preparing
  432. 22:01the files is easy. I've already given
  433. 22:03them an overview of how we receive the
  434. 22:05files
  435. 22:08and verify them.
  436. 22:10They'll write code to check this, and
  437. 22:12you can let them help build the
  438. 22:14components.
  439. 22:15Then add the necessary parts.
  440. 22:18The data files are organized to prevent
  441. 22:20duplication, using generated numbers.
  442. 22:25We're simply comparing the file
  443. 22:27codenames.
  444. 22:28Once that's done, we prepare for
  445. 22:30preprocessing.
  446. 22:32This means preparing the data for
  447. 22:34processing before starting the
  448. 22:36preprocessing process.
  449. 22:38This is called preprocessing–converting
  450. 22:41a PDF to an image, correcting image tilt
  451. 22:44, enlarging, sharpening, etc. After
  452. 22:46preprocessing, we call the model, send
  453. 22:49the image to the AI model to read, and
  454. 22:51then wait for the results.
  455. 22:55This is for the model to start reading;
  456. 22:57any model will do.
  457. 23:01And here, the table checking part is
  458. 23:03where we design the application–whether
  459. 23:06we want to check the table format, check
  460. 23:08the totals, etc. Wherever you want to
  461. 23:10highlight something in yellow or put a
  462. 23:12flag, this is how we design our harness
  463. 23:15.
  464. 23:17Actually, everyone on GitHub, especially
  465. 23:20those in the BoonBigNotePlus
  466. 23:24local group, can access it through the
  467. 23:26BoonBigNotePlus GitHub repository.
  468. 23:30For those in the community group
  469. 23:36here, you can download and use the
  470. 23:38application directly.
  471. 23:42Once you understand the basic principles
  472. 23:44, you can further develop this
  473. 23:46application.
  474. 23:49You can clone it from GitHub, fork it,
  475. 23:51or help develop it further on my GitHub
  476. 23:54repository so it can be included in the
  477. 23:56community group.
  478. 23:59This is regarding the distribution on
  479. 24:01GitHub.
  480. 24:03If anyone wants to use GitHub but
  481. 24:05doesn't know how, you can ask me on my
  482. 24:07profile.
  483. 24:10If you're in the SchoolBoonBigNotePlus
  484. 24:12group, you can contact me privately, and
  485. 24:14I'll migrate you free of charge.
  486. 24:18This is step 5; check the components,
  487. 24:21look at the yellow warning indicators,
  488. 24:23and save.
  489. 24:25It's simple. For exporting, you can
  490. 24:27choose whether you want an Excel file, a
  491. 24:30CSV file, or an SQLite file, or whether
  492. 24:32you want to save it on your computer.
  493. 24:34You can design it however you like.
  494. 24:37I've designed it to back up the data to
  495. 24:40SQLite and export to Excel as well.
  496. 24:43Let's look at the application's results
  497. 24:45.
  498. 24:46Okay everyone, this is the repository
  499. 24:48I'm talking about.
  500. 24:54For private communities, it includes VPS
  501. 24:56setup and, you know, the observe feature
  502. 24:59, which is for organizational use.
  503. 25:06I'll reveal it to everyone soon, but
  504. 25:08this is just a brief overview.
  505. 25:13Okay, everyone, here it is.
  506. 25:18It will include organizational chat, OCR
  507. 25:21login, and so on.
  508. 25:25Okay, so that's it.
  509. 25:29I'll be using this to download the
  510. 25:31models.
  511. 25:34It's like our OCR, the one with the Goji
  512. 25:37and Rama models.
  513. 25:40We can download the AI, it's open source
  514. 25:42.
  515. 25:45Okay, and we've already prepared for
  516. 25:47security, data hashing, and cipher
  517. 25:49implementation.
  518. 25:51This is about storing the key,
  519. 25:53specifically the file extension name,
  520. 25:55and backing up.
  521. 25:59This includes signing with crypto to
  522. 26:01protect the data.
  523. 26:05You can open the project in any disk or
  524. 26:07computer environment.
  525. 26:13The concept is the same: click "new,"
  526. 26:16then remember to open "diskcomputer" and
  527. 26:18"openfolder"(it's like opening a project
  528. 26:21, for example, the same thing).
  529. 26:29This is working within the same project
  530. 26:31, or even in a chat.
  531. 26:35Everyone follows the same principle.
  532. 26:39Once you're done, open
  533. 26:45the project, select "new project," and
  534. 26:47you might need to type a name.
  535. 26:53They're not all the same, I'll start
  536. 26:55with this much, then add more, and it
  537. 26:57will be integrated into this program.
  538. 27:00Everyone follows the same principle,
  539. 27:02regardless of the course.
  540. 27:04If you have any other programs, you can
  541. 27:07run them like this.
  542. 27:09Okay, then run it here.
  543. 27:11This is the program for those who
  544. 27:13already have it and want to run it.
  545. 27:16This is how to run it. But if it doesn't
  546. 27:18run, you can tell me.
  547. 27:20Just open the app normally.
  548. 27:24This is also something I've prepared for
  549. 27:26you.
  550. 27:29You can check the box or verify it
  551. 27:31before actually running this app.
  552. 27:34And this addresses common problems that
  553. 27:36many people may encounter.
  554. 27:39I've compiled them all. If you misread a
  555. 27:41lot, it might be due to your own
  556. 27:42understanding. Is it the source image?
  557. 27:44Why is it slow?
  558. 27:46Are we using a smaller model, or a
  559. 27:48larger one but there's not enough RAM?
  560. 27:51Or are we using a smaller model and can
  561. 27:54we adjust the resolution to 4-bit, or 4
  562. 27:56-bit?
  563. 27:58It's like a resolution level, something
  564. 28:00like that.
  565. 28:02This is some terminology.
  566. 28:08We also need to check if the input data
  567. 28:10is correct.
  568. 28:12This is a system.
  569. 28:15Okay, to summarize, this video covers
  570. 28:17everything from installing the Typhoon
  571. 28:20Orama, to
  572. 28:27creating security and running it in
  573. 28:29real-world mode.
  574. 28:33Today, I'm going to use Google
  575. 28:35Anti-Gravity.
  576. 28:39I recommend using the Google
  577. 28:40Anti-Gravity
  578. 28:46feature. Antigravity is quite high,
  579. 28:48everyone. Guild has
  580. 28:52this
  581. 28:57slash
  582. 29:02, and
  583. 29:07Guild
  584. 29:14has...
  585. 29:22okay?
  586. 29:28Okay,
  587. 29:39okay...
  588. 29:49Okay, I'll try selecting a PDF file
  589. 29:51first, which I've prepared.
  590. 29:54It might be in your download. Here it is
  591. 29:56, everyone, the invoice sheet. Let's
  592. 29:58take a look.
  593. 30:00Wow, everyone, it's here!
  594. 30:03And it's in fullscreen.
  595. 30:06Okay, 0%tilt, definitely, because this
  596. 30:09is a sample file.
  597. 30:13And it only took 200ms to find 38 lines
  598. 30:17.
  599. 30:20Wow, everyone, it has file encryption.
  600. 30:24Okay, regarding security.
  601. 30:27And the invoice is correct.
  602. 30:29It can also be autotable.
  603. 30:31Wow, it meets all the requirements. It
  604. 30:34seems like it knows that this is an
  605. 30:35invoice, and there's a sample invoice
  606. 30:37too.
  607. 30:38Wow, everyone, there's a sample invoice
  608. 30:41as well.
  609. 30:42OCR is running. It's already running.
  610. 30:45And we can save it to SQLite as well.
  611. 30:48You can export to Excel too, everyone!
  612. 30:51Wow, this program is perfect!
  613. 30:56Actually, everyone can download it from
  614. 30:58the private community, or for those
  615. 31:00already in the free community, it's
  616. 31:02available here.
  617. 31:07But I'll be developing this application
  618. 31:10on GitHub continuously; it's a CI&CD
  619. 31:12application.
  620. 31:15You can follow along and discuss it
  621. 31:17further.
  622. 31:19That concludes the learning in this
  623. 31:20video, but we can build upon it in the
  624. 31:22community.
  625. 31:26For those who really want to learn about
  626. 31:28AI and its real benefits, let's help
  627. 31:30develop a better society for Thailand
  628. 31:32and share good AI and good tools because
  629. 31:34we have them.
  630. 31:36The idea is already great, and we'll go
  631. 31:39global and sell it internationally.
  632. 31:41It's really not an exaggeration; this
  633. 31:43kind of thing only requires our
  634. 31:45cooperation, unity, and mutual support
  635. 31:48within this community.
  636. 31:51Okay, that's all for today.
  637. 31:55We'll see each other in the next episode
  638. 31:57or in the community.
  639. 32:00Goodbye for now.
  640. 32:01Oh, and don't forget, everyone, the
  641. 32:04Typhoon OCR model and OpenThai's work
  642. 32:06are amazing!
  643. 32:08And in the next video, I'll teach you
  644. 32:10how to train your own local AI if you
  645. 32:12already have data within your
  646. 32:14organization.
  647. 32:16Let's wait and see in the next video.
  648. 32:19Goodbye!

About this transcript

This page contains the full transcript of สร้างแอป Local AI แปลงไฟล์ PDF เป็น Excel ภาษาไทยไม่เพี้ยน! by BoomBigNose, generated from the public captions YouTube serves with the video. The transcript has 3,484 words across 648 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.