ROADMAP to Become an AI Engineer in 2026 (Beginner to Pro) — Transcript

Full transcript

1. 0:00Recently openai launched their chart
2. 0:01GPT6 or Astra model and everywhere on
3. 0:04the internet everyone is talking about
4. 0:05that AGI is here. AGI fullform
5. 0:08artificial general intelligence and it
6. 0:10is actually intelligent as a general
7. 0:13person. But the question that everyone
8. 0:15is asking is doing AI engineering right
9. 0:18now is a good choice or not? And
10. 0:19honestly, this is the best time to
11. 0:21become an AI engineer because at this
12. 0:23moment AI engineering is moving so fast
13. 0:26and if you are at the top tier, if you
14. 0:28are actually learning all the skills
15. 0:29that are required to become an AI
16. 0:31engineer, then companies are dying to
17. 0:34hire you. And every company in India, in
18. 0:37GCC, in US is trying to hire an AI
19. 0:40engineer because everyone wants to
20. 0:42integrate AI solutions into their
21. 0:43workflows. And if you don't know, I'm
22. 0:45Nishant. I averaged $250 US for my AI
23. 0:48startup and in that startup my
24. 0:50co-founder is an AI engineer. He used to
25. 0:52work at midlife and his CTC was around
26. 0:5550 lakh rupees. I've also tried to
27. 0:57answer all the questions that comes to
28. 0:59your mind when you are trying to
29. 1:00research for AI engineering roles and
30. 1:02when you are trying to research what you
31. 1:03want to do and you don't need to be a
32. 1:05math genius to become an AI engineer.
33. 1:07And the interesting part is the
34. 1:09opportunity is not limited to people who
35. 1:11are doing research or PhD. The problem
36. 1:13is most people don't know where they
37. 1:14have to start. They just think AI
38. 1:16engineering is hard mathematics,
39. 1:18research papers or they need a PhD and
40. 1:21you don't. In this video, we'll be
41. 1:23talking about what are the things you
42. 1:24need to learn to go from a beginner to a
43. 1:26job ready AI engineer. It also helps you
44. 1:29improve your ATS score for your resume,
45. 1:31create a cover letter, optimize your
46. 1:32LinkedIn profile and everything can be
47. 1:34done in few clicks. I would highly
48. 1:36recommend you to check out hackats.com.
49. 1:38We are adding so many features [music]
50. 1:39every single day so that we can make
51. 1:41your job finding process easy. Hi guys,
52. 1:43I'm Nishan Zahar. Welcome back to the
53. 1:44channel. I'm not going to waste time by
54. 1:46giving my introduction. You can check
55. 1:47out that anywhere. Let's just start with
56. 1:50the road map. First, we need to
57. 1:51understand what an AI engineer actually
58. 1:53does because most people don't actually
59. 1:55know what an AI engineer does. They get
60. 1:57confused between a machine learning
61. 1:58engineer and an AI engineer. By looking
62. 2:00at the name, people think that AI
63. 2:02engineers are responsible for creating
64. 2:04models or training models. But that part
65. 2:06is actually closer to a machine learning
66. 2:08engineer or a researcher. What an AI
67. 2:11engineer does is they take the models
68. 2:12that are already there. They use them to
69. 2:14create real world solutions for problems
70. 2:16that exist around us. That means they
71. 2:18don't just play around with the model.
72. 2:20They actually create build applications,
73. 2:23publish them, deploy them and actual
74. 2:26users use them. And you need software
75. 2:28engineering fundamentals because
76. 2:32then you'll not be able to integrate AI
77. 2:34properly. AI should not feel like a
78. 2:36feature that you have added on top of it
79. 2:37when you are creating any AI
80. 2:38application. It should be integrated in
81. 2:41the workflow so that the user doesn't
82. 2:43even know that AI was used or there is
83. 2:45AI inside the product. So let's start
84. 2:47with the first layer which is software
85. 2:48engineering fundamentals that you need
86. 2:50to learn. Now because AI engineers don't
87. 2:52just have to use LLMs they actually have
88. 2:54to build real world software. You need
89. 2:56to learn basics of software engineering
90. 2:58as well. You don't have to go in depth.
91. 3:00First you need to learn a programming
92. 3:01language. Whenever you do anything
93. 3:03technical, you need to know a
94. 3:04programming language for sure. In this
95. 3:06case, we are going to learn Python. The
96. 3:07most popular libraries and frameworks
97. 3:09that we use in machine learning are in
98. 3:11Python. And that is why Python should be
99. 3:13your primary language. I've already
100. 3:15created a dedicated video on how you
101. 3:17should learn Python. What is the road
102. 3:18map for that? I won't spend time on
103. 3:20that. There the video is in the I
104. 3:21button. You can definitely check that
105. 3:22out. Second thing is API. API is
106. 3:25application programming interface. That
107. 3:27means how two softwares or two services
108. 3:29actually interact with each other. That
109. 3:31is what an API is. You need to
110. 3:33understand how these APIs are created,
111. 3:35how you can create this, how you can
112. 3:36interact with the APIs and how you
113. 3:39actually use them in the production
114. 3:40environment. This is also important for
115. 3:42an AI engineer because you have to use
116. 3:44the AI model that you use in your
117. 3:47products through APIs. Third thing is
118. 3:49going to be git, GitHub and basic cloud.
119. 3:51Get GitHub because you are going to work
120. 3:53in an environment with other engineers
121. 3:55as well. So you need to learn basics of
122. 3:57Git and GitHub. It is used for version
123. 3:59control. Next is going to be cloud
124. 4:00because you are going to deploy your
125. 4:03application finally on some cloud
126. 4:05server. That's why you need to learn
127. 4:06cloud as well. So I've already created a
128. 4:08dedicated road map for Python. You can
129. 4:11definitely check that out. All the
130. 4:12resources are mentioned there. I'll also
131. 4:13mention them in the doc. Second, if you
132. 4:15want to learn about git, you can check
133. 4:17out this video by free code camp. And
134. 4:19even if you want to learn about APIs,
135. 4:21this is also another video by free code
136. 4:24camp on APIs that you can check out.
137. 4:25There is a 80/20 rule. If you want to
138. 4:27learn 80% of the things in a programming
139. 4:29language, you just need 20% of the time.
140. 4:31So if you want to learn Python and all
141. 4:33these things, it should not take you
142. 4:35more than 3 to 4 weeks if you have some
143. 4:37background in software engineering. Even
144. 4:39if you don't have, it won't take you
145. 4:40more than 6 weeks. Now comes the layer
146. 4:42two which is going to be AI and LM
147. 4:44fundamentals. The most important part
148. 4:46here is that you don't need to
149. 4:48understand deep mathematics. You don't
150. 4:50need to understand how the actual
151. 4:52transformers or your attention matrices
152. 4:55work. You just need to have the basic
153. 4:57idea on how these things are built. No
154. 5:00one is going to ask you to actually use
155. 5:02them or do the maths behind them. That
156. 5:04level of understanding is more relevant
157. 5:06for researchers or ML engineers. You
158. 5:08just need to understand the practical
159. 5:09intuition behind on how LMS actually
160. 5:11work. And at the basic level, these are
161. 5:13the three things that you need to know.
162. 5:15The first is tokens. What do models do?
163. 5:17They break the text that you are talking
164. 5:18about into multiple set of tokens. And
165. 5:21that is why the length of the prompt
166. 5:22that you are giving actually matters
167. 5:24when you're calculating the cost. Second
168. 5:25is going to be temperature. It controls
169. 5:27how random or predictable your output is
170. 5:29going to be. If you change the
171. 5:30temperature, it is possible that the
172. 5:32output for the same prompt can be
173. 5:34different if you increase the
174. 5:35temperature, decrease the temperature or
175. 5:37you do anything with that. And third is
176. 5:38going to be knowing which model is good
177. 5:40for what. And this can only be done when
178. 5:42you actually experiment with things. And
179. 5:43you won't be able to understand all this
180. 5:45by only watching some tutorials or
181. 5:47actually just reading articles. You need
182. 5:49to get your hands dirty. Pick an AI
183. 5:51model, go to charg or claude and
184. 5:54actually run the same prompt with
185. 5:55different temperatures. You can go to
186. 5:56platform.openai.com and you will find
187. 5:59all these settings there. Also, there is
188. 6:01an amazing video by Andre Karpathi on
189. 6:02intro to large language models. This is
190. 6:05one of the best videos that you can find
191. 6:07on how LMS actually work. Do check it
192. 6:09out. Do watch that full. Second, if you
193. 6:11want to understand how tokens actually
194. 6:12work and how text is broken down into
195. 6:14context, what you can do is you can go
196. 6:16to this website which is
197. 6:16platform.opi.com/ openi.com/to
198. 6:19tokenizer. Whatever prompt you add, it
199. 6:21will show you how and what are the exact
200. 6:24tokens and how your prompt is being
201. 6:26divided into multiple tokens. This layer
202. 6:28is relatively small. You should not take
203. 6:29more than 1 to two weeks in finishing
204. 6:31this. Layer three is going to be core AI
205. 6:33implementation. Once your basics are
206. 6:34done, now you need to actually build
207. 6:37stuff. And here there are two concepts
208. 6:39which are absolutely important for you.
209. 6:41The first one is prompting. It's not
210. 6:42just about asking the smart questions.
211. 6:44It's about giving the model clear
212. 6:46instructions. so that you get a
213. 6:47predictable and reliable output. There
214. 6:50are three things in this. The first one
215. 6:51is context window. So what do you mean
216. 6:53by context window? So context window is
217. 6:55the amount of information or the memory
218. 6:57a model can save. Earlier when charge
219. 6:59GPT the initial charge GPT models came
220. 7:02the context window was pretty small. It
221. 7:04was less than 1,000 2,000 words and
222. 7:06that's why these models were
223. 7:07hallucinating. As the models got better,
224. 7:10the context window improved which has
225. 7:12more than 1 million tokens context
226. 7:13window. Second is going to be system
227. 7:15prompts and third is going to be
228. 7:16temperature. As we talked about this is
229. 7:18the control which decides how
230. 7:19predictable the output is going to be or
231. 7:21how creative it is going to be. And one
232. 7:23more important skill that you need to
233. 7:24learn is how to get structured outputs.
234. 7:27Most of the outputs that we get are in
235. 7:28the JSON format because at the end you
236. 7:31want the data to be used somewhere. If
237. 7:33you don't get the output in a structured
238. 7:34format, you'll not be able to show it to
239. 7:36the user. Now the next and the most
240. 7:38important part in the whole AI
241. 7:40engineering thing is rag because most of
242. 7:43the companies I I would say all of the
243. 7:45companies need someone who can implement
244. 7:46rag for their systems so that whatever
245. 7:49the output is by the LM by the model is
246. 7:52actually relevant to their customers to
247. 7:54their team or any of the stakeholders.
248. 7:57What is rag now? RAG is retrieval
249. 7:59augmented generation. Now think of it if
250. 8:01you work at Adobe there are so many
251. 8:02instructions so many docs which the LLM
252. 8:05is not trained on. If you have a big
253. 8:07code base there are so many dogs that
254. 8:10would have been written. If you want to
255. 8:11create a chat bot so that other teams
256. 8:13can also understand what you have done.
257. 8:15There is no direct way to use the same
258. 8:18data for training your model. The only
259. 8:21way was either fine-tuning or training
260. 8:22the model. But in that way it is a
261. 8:25longer way and you don't want to do
262. 8:27that. Now how do you actually give
263. 8:28context to a model and for this you use
264. 8:31rag. There are different types of rags
265. 8:33and there is a whole like you need to
266. 8:36get into the depth of it. There is graph
267. 8:38rag, there is your normal rag, then
268. 8:40there is aentic rag, there is vector
269. 8:42search. There are so many things that
270. 8:43you need to learn in this. Uh I'll just
271. 8:45talk about the basics but you need to go
272. 8:47in depth. I'll give you resources where
273. 8:49you can actually learn in depth. For
274. 8:51prompting you can check out the free
275. 8:52official guides by enthropic or openai.
276. 8:54Both the links are in the description
277. 8:55and you can see them here as well. For
278. 8:57rag, you can check out the video uh rag
279. 9:00from scratch by free code camp. This is
280. 9:02important the basics. It will take you 3
281. 9:04to 4 weeks but you need to go in depth
282. 9:07as well. So add another month for that
283. 9:09when you are building projects. You can
284. 9:11try out different types of rags for each
285. 9:13of this and you can actually showcase
286. 9:15that as well. Fourth is going to be
287. 9:16advanced agentic systems. Till now the
288. 9:19model was only able to answer you right
289. 9:21you were asking some questions and the
290. 9:23model was answering. Now the next part
291. 9:25is actually getting some work done by
292. 9:27these models and that is where the
293. 9:29concept of agent actually comes. So the
294. 9:31difference between an AI agent and a LLM
295. 9:34or a model is that AI agent has some
296. 9:36tools that it can access. [music] It can
297. 9:38be a browser. It can be a video editor.
298. 9:40It can be cloud like a AWS or Azure. It
299. 9:44has access to multiple tools which it
300. 9:46can invoke whenever required. An AI
301. 9:48agent can actually take actions as well
302. 9:49like calling an API, [music] invoking,
303. 9:51triggering a workflow, going through the
304. 9:53database, getting some data from
305. 9:55database, understanding that giving you
306. 9:57the final output like all these things
307. 9:58can be done by an AI agent. And most of
308. 10:01the enterprises right now want a mixture
309. 10:03of these agentic rag systems because
310. 10:06they want uh to create their rag and
311. 10:09agents that can do a lot of stuff. And
312. 10:11to get the work done by an agent, there
313. 10:13are two things which are absolutely
314. 10:14important. The first one is function
315. 10:16calling and the second one is
316. 10:17frameworks. What do you mean by function
317. 10:19calling? So basically once the AI model
318. 10:21decides what are the tools it wants to
319. 10:23call they [music] it needs to call these
320. 10:25functions as well. Needs to call these
321. 10:27tools as well and this is called
322. 10:28function calling. Second is [music]
323. 10:29frameworks. When you are doing
324. 10:30multi-step task or orchestration there
325. 10:33are few tools that you need to learn to
326. 10:35get these things done. And there is one
327. 10:36more concept that you need to learn
328. 10:38which is MCP which is called model
329. 10:39context protocol. Think of MCP as a way
330. 10:42to talk to any tool that was built
331. 10:45before the AI era. If I want to talk to
332. 10:48AWS, there are there is an MCP server
333. 10:51which has a list of tools it can call.
334. 10:52My LLM can call and this is what an MCP
335. 10:55server is where they have given you a
336. 10:57list of tools that you have access to.
337. 10:59Your LLM has access to and they can call
338. 11:02it. It is an advanced topic. So
339. 11:03initially you need to just learn the
340. 11:05basics and when you use it you can go in
341. 11:07depth. Here are few free resources that
342. 11:09you can definitely check out. The first
343. 11:10one is by Microsoft AI agents for
344. 11:12beginners course. This is on GitHub.
345. 11:15This is free on GitHub. It has 12
346. 11:16lessons. And here you'll be learning
347. 11:18agents tools and multi- aent systems.
348. 11:21[music] And second is going to be AI
349. 11:22agent course by hugging face. Here
350. 11:24you'll be learning how you can create
351. 11:25agents. You can [music] build agents.
352. 11:27And all this is absolutely free. You
353. 11:29also get a certification. That is not
354. 11:31the important part. But yeah, this whole
355. 11:33step is going to take you anywhere from
356. 11:343 to 4 weeks to get [music] the gist of
357. 11:37it. Now I understand this sounds like a
358. 11:39lot of new terminologies. MCP agent
359. 11:41multi- aent orchestration rag and like
360. 11:44so much more. So it can be confusing for
361. 11:46sure. For this only I found a really
362. 11:48interesting tool. This is called explain
363. 11:50and it is built by scrimbar. Let me show
364. 11:52you. So I've just asked what is MCP and
365. 11:54AI? I'll ask it to explain and it will
366. 11:57create an explainer for me.
367. 12:00>> MCP stands for model context protocol
368. 12:02and it's basically a universal plug for
369. 12:04AI. Think of MCP like a USBC port for
370. 12:08AI.
371. 12:08>> As you can see, it is generating
372. 12:09everything real time for me. Initially
373. 12:12when we started it was only 7 seconds
374. 12:14and now it's 1 minute. It gives you a
375. 12:16clear video explanation. There is no
376. 12:17unnecessary jarens and it comes in 33
377. 12:20languages so that it gives you more
378. 12:22accessibility as you can see
379. 12:25GPS.
380. 12:31And if I want to go deeper, I want to
381. 12:33understand the concept more, what I can
382. 12:35do is I can actually click on go deeper
383. 12:37and ask follow-up questions as well. And
384. 12:39if I want to learn the topic in more
385. 12:40detail, what I can do is I can even
386. 12:42convert this into a complete mini course
387. 12:44or a playlist. You can even install
388. 12:45their Chrome extension and you can
389. 12:46understand any difficult topic. If
390. 12:48you're reading a research paper, an
391. 12:49article, you can just go there and ask
392. 12:51it to explain the same to you. It will
393. 12:53break the article down into simpler and
394. 12:55easier to understand parts. You can also
395. 12:57bring this into your existing workflow.
396. 12:59Explain works with claude, cursor,
397. 13:01codeex through MCP and it also has a
398. 13:03chart GPT plug-in. So whenever you get
399. 13:05stuck on a new concept, you can actually
400. 13:07use this as your personal tutor. Having
401. 13:09something like this can actually make
402. 13:11your process so much easier. And that's
403. 13:13why I found explain pretty useful. If
404. 13:15you want to try it out, the link is in
405. 13:17the description. Now let's come back to
406. 13:18the road map. Now comes the layer five.
407. 13:20Now you have already created the AI
408. 13:22application. You learned how to create
409. 13:23agents. [music] You learned how to add
410. 13:25drag. You learned most of the things.
411. 13:27Now the main part is the most important
412. 13:30problem that all of us have seen in our
413. 13:34past that it worked on my machine and it
414. 13:36is not working on the production
415. 13:37environment and to solve this we have
416. 13:39docker and that is why you need to learn
417. 13:41the basics of mlops as well you can call
418. 13:44it mlops you can call it a ops so first
419. 13:46you need to learn docker for this it
420. 13:48converts your entire application into a
421. 13:50container so it runs consistently across
422. 13:52different platforms across different
423. 13:54machines so whether it's on your laptop
424. 13:56or it's on cloud it will actually work.
425. 13:58Second is going to be cloud platforms
426. 14:00like Azure, AWS, GCP whichever one you
427. 14:02like. This is where you are going to
428. 14:03host your application. So you need to
429. 14:05learn the basics of this as well. And
430. 14:07third is going to be cost control
431. 14:08because everyone wants the cost to be
432. 14:10down and you need to know how you can
433. 14:12bring the cost down not just by reducing
434. 14:15the model that you are using or reducing
435. 14:17the intelligence of the model that
436. 14:18you're using but actually doing
437. 14:20optimizations throughout your process
438. 14:22throughout your pipeline. And if you can
439. 14:24reduce the cost for any company that
440. 14:26then you are going to be really valuable
441. 14:28for that. If you want to learn Docker,
442. 14:30there's a free course by a free code
443. 14:31camp. You can check that out. If you
444. 14:33want to learn about LLM ops, there is a
445. 14:35course on deep learning AI that you can
446. 14:38check out here. This should take you
447. 14:40anywhere from 3 to 4 weeks. Now the most
448. 14:42important part, the most important layer
449. 14:44is going to be your career strategy
450. 14:47because the things in the AI world are
451. 14:49changing so fast. There are so many
452. 14:51things that are going on and so many new
453. 14:52things that are coming up that you need
454. 14:54to stay updated. When you are actually
455. 14:56learning how to become an AI engineer,
456. 14:57the last thing that you need to do is
457. 14:59adding the projects that are relevant
458. 15:00for becoming an AI engineer in your
459. 15:02resume. If you just say that I am an AI
460. 15:05engineer, I've done these
461. 15:06certifications, no one is going to
462. 15:07actually care what what you have done
463. 15:09until and unless you can actually show
464. 15:11them what you have done. So to show them
465. 15:13you need to create real world projects
466. 15:14or you need to do some internships. You
467. 15:16need to create products that you have
468. 15:17shipped, users are using. That is what
469. 15:20you need to do in 2026. So if you are
470. 15:23already a software engineer, uh I told
471. 15:25you in the start of the video about hack
472. 15:26ATS, we have built a product. There is
473. 15:28one more feature in that which I find
474. 15:31very very very interesting. So it is
475. 15:33around how you can add projects in your
476. 15:36resume so that you can show that you
477. 15:38have learned these skills as well. So
478. 15:41what you have to do is you just have to
479. 15:42upload your resume. It will ask you some
480. 15:45questions on which roles you want to
481. 15:47target. You can tell that it will find
482. 15:49all the jobs that are relevant for you
483. 15:52uh and [music] give you a ATS score.
484. 15:54Give you a relevancy score. After
485. 15:55getting the relevancy score, you can
486. 15:57actually check out the ATS score for
487. 15:59your resume. If the ATS score is low, it
488. 16:01will recommend you the courses that you
489. 16:03can do for the cluster of jobs that you
490. 16:06can apply to. Once you do that, you can
491. 16:07actually improve your ATS score as well.
492. 16:09and it will just give you a step-by-step
493. 16:11guide on how to build it. It is going to
494. 16:13be very relevant to you if once you do
495. 16:16that and give it the final GitHub repo.
496. 16:19It will give you the final pointers that
497. 16:22you need to add in your resume so that
498. 16:24your resume is again ATS friendly and
499. 16:26you can apply to any job that you want.
500. 16:28It will tailor your resume for each of
501. 16:30the jobs that you want to apply to. So
502. 16:32there are free credits that I'm giving
503. 16:34to people who are going through this
504. 16:36video. So do check it out and for any
505. 16:39feedback there is a support email that
506. 16:41you can email us to. We would love to
507. 16:43know what all new features you want in
508. 16:45this product. Apart from that there is a
509. 16:47GitHub template on how you should create
510. 16:50the readme file, how you should add the
511. 16:52video on uh on whatever product that you
512. 16:55have built. I've also attached all that
513. 16:56in the document and in the description.
514. 16:58So you should check that out too. Now
515. 17:00the last part and the most interesting
516. 17:01part is the salary and the actual
517. 17:04reality. So if you are a fresher who
518. 17:06wants to become an AI engineer and this
519. 17:08is going to be your first role, you can
520. 17:10expect anywhere from 12 lakhs to 15
521. 17:12lakhs in India. If you come to the Gulf
522. 17:15area, if you're able to find a job in
523. 17:17GCC, if you are find able to find a job
524. 17:18in Europe or outside India, the salaries
525. 17:20are going to be crazy. In GCC that means
526. 17:23Dubai and the Gulf area, the salaries
527. 17:25can go anywhere from 50 lakh rupees to 1
528. 17:28K. Once you have 2 to 4 years of
529. 17:30experience, your salary can go anywhere
530. 17:32from 24 lakhs to 50 lakh rupees. My
531. 17:34co-founder had a three years of
532. 17:35experience and his salary was in that
533. 17:38range. He was he's from IIT Del. So that
534. 17:40also plays some role but if some if you
535. 17:42are not from a very prestigious college
536. 17:44but you have done an amazing projects
537. 17:47you can [music] get into YC companies
538. 17:48you can get into startups and create
539. 17:50your portfolio and not just salary range
540. 17:52these are the skills which are really
541. 17:54very important if you're applying for AI
542. 17:56engineering roles and the skills are
543. 17:58generative ILM fine tuning MLOps AI
544. 18:00agents these are the skills that you
545. 18:02need to learn. We did not talk about LLM
546. 18:04finetuning in this whole video because
547. 18:07it it would have taken more time and
548. 18:09there are some nuances that you need to
549. 18:11discover yourself the things that you
550. 18:13need to find yourself to become a good
551. 18:15AI engineer and a good engineer in
552. 18:17general. In 2025 there were around 2.9
553. 18:20lakh job posting for AI engineering
554. 18:22roles. In 2026 it's around 3.82 lakhs.
555. 18:25There's a 32% jump in just one year and
556. 18:27this is going to increase. So this is
557. 18:29the skill that you need to learn in
558. 18:302026. If I talk about the timeline, it
559. 18:32can take you anywhere from four months
560. 18:34to 6 months and even more depending upon
561. 18:36how much time you are actually giving to
562. 18:38learn this. But realistically, it will
563. 18:40take you 8 to 12 months to actually
564. 18:41become an AI engineer. So that is it.
565. 18:43Next step is on you. You need to start
566. 18:46learning today. You need to build
567. 18:47projects. Do check out scrimar. Do check
568. 18:49out hackats whenever you are applying
569. 18:51for jobs. And thank you for watching
570. 18:52this video. I hope I was able to help
571. 18:54you out. If you like this video, do
572. 18:55click on the like button. Do subscribe
573. 18:57to the channel. And I'll see you in the
574. 18:58next one. Bye.