AWS AI/ML & Analytics Explained | SageMaker, Bedrock, Athena, Glue & More | AWS CLF-C02 Day 19 — Transcript

Full transcript

1. 0:01Hey, welcome back. Day 19 is here. We
2. 0:04are in the home stretch now. Only two
3. 0:05more after this. Today we are talking
4. 0:08about something that honestly makes AWS
5. 0:10feel a little futuristic. AI and machine
6. 0:12learning. Plus how AWS handles massive
7. 0:15amounts of data. Think about it. Every
8. 0:17time Netflix recommends a show or bank
9. 0:20flags a weird transaction or let's say a
10. 0:22chatbot answers your question instantly,
11. 0:24there's a service behind that doing the
12. 0:26heavy lifting. Today you will learn the
13. 0:28AWS versions of exactly that. Now here's
14. 0:32the good news for your exam. You are not
15. 0:34going to be asked to build any of this.
16. 0:36Nobody's going to ask you to train a
17. 0:38neural network. The exam just wants to
18. 0:40know can you match the right AWS service
19. 0:42to the right business problem or not.
20. 0:44That's it. So today is really a matching
21. 0:46game and once you see a pattern, it
22. 0:48clicks fast. So let's get into it. So
23. 0:51quick road map for today. We are going
24. 0:53to cover this in a very natural order
25. 0:55like a story. First we will zoom out and
26. 0:57understand the big picture of AWS AI
27. 0:59services. What's ready versus what you
28. 1:02build yourself. Then we will go service
29. 1:04by service through the readymade AI
30. 1:06tools. First the ones that deal with
31. 1:08images, documents and language. And then
32. 1:11the ones that deals with speech,
33. 1:13translation, chats bots and search.
34. 1:16After that we shift gears into
35. 1:17analytics. How AWS moves data from raw
36. 1:19and messy all the way to the polish
37. 1:21dashboard someone can actually use to
38. 1:23make decisions. and then we will wrap up
39. 1:25with the quick reference table and have
40. 1:27a fun challenge as well. So, sounds
41. 1:30good. Let's start with the big picture.
42. 1:33Okay, imagine you need a translator. You
43. 1:36have two options in real life. You
44. 1:38either hire someone who already speaks
45. 1:4015 languages fluently and then just ask
46. 1:42them to translate your document right
47. 1:43now or you train someone from scratch to
48. 1:46become translator which takes way longer
49. 1:48but gives you something incredibly
50. 1:50specific to your needs. That's basically
51. 1:53the difference between AWS two
52. 1:54categories of AI services. So we have
53. 1:57readym made AI services. They are like
54. 2:00the fluent translator who is already
55. 2:02sitting there. AWS has already trained
56. 2:04the model on huge amounts of data. You
57. 2:06just call an API meaning you send a
58. 2:08request and get a answer back. No
59. 2:11machine learning knowledge is required.
60. 2:12It is fast, easy, but you can't
61. 2:14customize how it thinks. So we have AWS
62. 2:18SageMaker AI. It is the train someone
63. 2:20from scratch option. It's a full
64. 2:22platform for building, training, and
65. 2:24deploying your own custom models for
66. 2:26when your problem is so specific that no
67. 2:28readymade service fits. Maybe you want
68. 2:30to predict which of your customers will
69. 2:32cancel their subscription based on your
70. 2:34company's unique data. That's a
71. 2:35SageMaker AI job. Here's the rule of
72. 2:37thumb. I really want you to lock this in
73. 2:40because exam loves to test this exact
74. 2:42idea. If the task is common, translate
75. 2:44text, transcribe a voice recording,
76. 2:47detect object in a photo, that's a
77. 2:49readymate service. If it's a custom
78. 2:52prediction model trained on company's
79. 2:53own proprietary data that's sagemaker
80. 2:55AI. Now two more names you will bump
81. 2:58into Amazon Bedrock and we have Amazon
82. 3:01Q. So Amazon Bedrock is AWS platform for
83. 3:04building generative AI applications.
84. 3:07Think of chatbots or content generators
85. 3:09using powerful foundational model from
86. 3:11Amazon other providers without you
87. 3:12having to manage any servers behind it.
88. 3:15And then we have Amazon Q. It is an AWS
89. 3:18own AI assistant. It can answer
90. 3:19questions about AWS itself. It helps you
91. 3:22write code if you're a developer or
92. 3:23summarize business data depending on
93. 3:25which version you are using. You don't
94. 3:26need to memorize how these work
95. 3:28internally. Just recognize the names and
96. 3:30roughly what bucket they fall into. So
97. 3:32let's meet the readymade services one at
98. 3:34a time because honestly once you know
99. 3:36what each one's superpower is, this
100. 3:37becomes really easy. So the first one we
101. 3:40have Amazon recognition. Think of the
102. 3:42word recognition. It looks at images and
103. 3:45videos. It can detect faces, compare two
104. 3:47faces to see if they match, recognize
105. 3:49objects and scenes, and even flag
106. 3:51inappropriate content. If the question
107. 3:53is about a photo or video, recognition
108. 3:55is your answer. Remember that. The next
109. 3:58we have is Amazon Textract. This one
110. 4:01pulls text and data out of scanned
111. 4:03documents. Picture a messy scanned
112. 4:05invoice or a form. Text track doesn't
113. 4:08just read the words. It understand
114. 4:10tables and fields. So, it knows this
115. 4:12number is the total and this box is the
116. 4:14date. That is smarter than basic OCR.
117. 4:17Now, here's a trap the exam loves. If
118. 4:19the question says, "Extract the total
119. 4:21from a scanned invoice," your brain
120. 4:23might jump into recognition because it
121. 4:25sounds like image stuff. But no, that's
122. 4:28extracted because it's a document with
123. 4:30structured information. Recognition is
124. 4:33for understanding what's in a photo or
125. 4:35video, not for reading structured text
126. 4:37out of a form. So, keep those separate
127. 4:39in your head. Then, we have Amazon
128. 4:42comprehend. This is about understanding
129. 4:44language and meaning in text. Sentiment
130. 4:46analysis. Is this review positive or
131. 4:49negative? Pulling out key phrases,
132. 4:51identifying names of people or companies
133. 4:53and even detecting what language
134. 4:55something is written in. An Amazon
135. 4:57comprehend medical is the specialized
136. 4:59cousin. It reads clinical notes and
137. 5:01pulls out things like medical
138. 5:03conditions, medication, and dosages.
139. 5:06Healthcare specific version of
140. 5:08comprehend. Now, let's cover the ones
141. 5:10dealing with voice, language,
142. 5:12conversation, and search. Amazon Poly
143. 5:15turns text into speech. Natural lifelike
144. 5:17audio. Think of an app that reads
145. 5:19article out loud to you. We have Amazon
146. 5:22Transcribe. It does the exact opposite.
147. 5:24Speech to text. You give it an audio
148. 5:27recording, it gives you a written
149. 5:28transcript, and it can even tell
150. 5:30different speakers apart in the same
151. 5:32recording. The next we have Amazon
152. 5:35Translate. Pretty self-explanatory.
153. 5:38Realtime translation between languages
154. 5:40keeping the meaning and formatting
155. 5:42intact.
156. 5:44Amazon lacks. It builds chat bots and
157. 5:46voice assistants. Fun fact, this is
158. 5:49literally the same technology behind
159. 5:50Alexa. It understand what someone means,
160. 5:53not just the exact word they typed,
161. 5:54which is called understanding intent.
162. 5:57And the last we have is Amazon Kendra.
163. 6:00It is enterprise search but smart
164. 6:02search. Employees can type a natural
165. 6:04question like what's our vacation policy
166. 6:05and kinder searches across internal
167. 6:07documents to find the answer instead of
168. 6:09you scrolling through folders. Here's a
169. 6:11classic trap. If say build a customer
170. 6:14service chatbot that's less because it's
171. 6:18conversation.
172. 6:20Let employees search through company
173. 6:22documents in plain English. That's
174. 6:24kendra because it's search not
175. 6:26conversation. Same vibe different job.
176. 6:29So don't mix them up.
177. 6:31All right, let's shift from AI to
178. 6:33analytics because these often get
179. 6:35grouped together on the exam, but they
180. 6:37are really about a different problem.
181. 6:40How do you take huge piles of raw data
182. 6:42and actually make sense of them? Think
183. 6:45of it like a factory line. Data comes in
184. 6:47raw. It needs to be clean and organized
185. 6:50and eventually someone needs to be able
186. 6:52to ask questions of it. Here, Amazon
187. 6:54Kinesis is step one, ingesting data in
188. 6:57real time as it happens. Think of
189. 7:00website clicks streaming in every second
190. 7:02or sensors on factory equipment
191. 7:03constantly sending readings. Kinesis
192. 7:06catches that live flowing data. And we
193. 7:09have AWS glue. It is a cleanup crew. It
194. 7:12is serverless ETL service which stands
195. 7:15for extract transformation load.
196. 7:19It takes up messy raw data and cleans
197. 7:21and prepares it. And glue also has a
198. 7:24neat side benefit. It can catalog your
199. 7:26data basically creating a map of what
200. 7:28data exist and where. So the services
201. 7:31like Athena can find and query it
202. 7:33easily. Talking about Athena, it is
203. 7:36where it gets really cool. Ethna lets
204. 7:38you run SQL queries meaning you can
205. 7:40literally ask questions like show me all
206. 7:42sales over $100 directly on the data
207. 7:45sitting in Amazon S3. No database to set
208. 7:47up, no servers to manage. You just pay
209. 7:49the amount of data your query actually
210. 7:52scans. So remember this if the exam
211. 7:54specifically mentions no servers to
212. 7:56manage query data sitting in S3 that is
213. 8:00almost always pointing you straight at
214. 8:01ethna. Let's continue the pipeline into
215. 8:04the bigger heavier tools. Amazon red
216. 8:06shift it is a full data warehouse built
217. 8:09for large scale analysis of structured
218. 8:12and historical data especially when you
219. 8:14need complex joins across huge tables.
220. 8:17Think of a company analyzing 5 years of
221. 8:19sales data.
222. 8:21And we have Amazon EMR. Amazon EMR runs
223. 8:25big data processing frameworks like
224. 8:27Apache, Spark and Hadoop on manage
225. 8:29clusters. This is for massive
226. 8:31distributed processing jobs. The kind of
227. 8:33scale you are crunching terabytes or
228. 8:35pabytes of raw data at once. The another
229. 8:38service we have Amazon open search
230. 8:40service. It is all about search and log
231. 8:43analytics at scale. Full text search,
232. 8:45monitoring application logs, building
233. 8:47operational dashboards to catch issues
234. 8:49fast. This is open search service.
235. 8:53And we have Amazon Quicksite. And this
236. 8:56is the final stop. A business
237. 8:58intelligence tool for building
238. 9:00interactive dashboards. This is what an
239. 9:02executive looks at. Clean charts and
240. 9:04visuals, not raw data. And here's the
241. 9:07trap that trips up almost everyone at
242. 9:08first. Athena and Red Ship both feel
243. 9:11like SQL databases. But Athena is
244. 9:14serverless. You pay per query straight
245. 9:16on data sitting in the S3. and we have
246. 9:19Red Shift. And Red Shift is a
247. 9:21provisioned cluster. You're paying for a
248. 9:23warehouse that's always running. So
249. 9:25don't swap these two on the exam. Make
250. 9:28sure to remember both of these services.
251. 9:30Let's pull this all together into one
252. 9:32simple reference because honestly this
253. 9:34table alone could save you points on the
254. 9:35exam day. Query data in S3 with no
255. 9:39database to manage that is Athena. Need
256. 9:42a dashboard for executives? We have
257. 9:45Quicksite.
258. 9:47And if you need of data warehouse for
259. 9:49structured historical data with complex
260. 9:51joins, we use Amazon Red Shift.
261. 9:55And if you need distributed big data
262. 9:57processing, we have Amazon EMR.
263. 10:01If need realtime streaming injection, we
264. 10:05use Kinesis.
265. 10:07For full text search or log analytics,
266. 10:10there is open search. And if you need to
267. 10:13clean or catalog data, we have glue.
268. 10:16Take a note, take a screenshot of this
269. 10:18one seriously because this kind of table
270. 10:20is you want to open the night before
271. 10:22your exam. Let's recap today in one bit.
272. 10:25Readymate AI means you just call an API.
273. 10:28AWS already trained it. SageMaker AI is
274. 10:31for building your own custom model and
275. 10:33Bedrock is for building generative AI
276. 10:36apps using foundational models and
277. 10:38Amazon Q's own AWS AI assistant.
278. 10:42And for vision and documents, Amazon
279. 10:44recognition handles videos and images.
280. 10:47Textract handles documents. For
281. 10:49language, comprehend understand text and
282. 10:52comprehend medical understands clinical
283. 10:54text. For speech and conversation, Poly
284. 10:56speaks transcribe listens and writes it
285. 10:58down. Translate switches languages. Lex
286. 11:01builds chatbot and Kendra powers smart
287. 11:03search. And for the data pipeline,
288. 11:06Kynesis brings data in. Glue cleans and
289. 11:08catalogs it. and Athena or Red Shift
290. 11:10lets you cur it and EMR handles massive
291. 11:13processing jobs. We have open search
292. 11:15that handles search and logs and
293. 11:16quicksite turns it all into a dashboard
294. 11:19someone can actually use. So tonight I
295. 11:21want you to flashcard this entire
296. 11:23service to task list and I also want you
297. 11:25to complete the assessment provided in
298. 11:27the LMS. This domain tends to show up as
299. 11:30three to five direct recall questions.
300. 11:32So this is genuinely free points if you
301. 11:34know your matching game. Tomorrow day 20
302. 11:37we move into migration strategies. the
303. 11:38well architected framework and the cloud
304. 11:41adoption framework and disaster
305. 11:42recovery. Big picture strategic AWS
306. 11:45thinking. See you there.