Hey, good morning, everybody. This is Julien from Arcee. I'd like to welcome you to the first episode of my podcast. I guess 2020 is just around the corner, so this is my New Year's resolution. In this podcast, we'll talk about the new announcements from AWS on AI and machine learning. I'll do some quick demos, and in the next episodes, I'll have guests who will explain what they build on AWS. By the way, if you're building cool stuff on AWS with machine learning, please get in touch and I'd love to have you on the podcast. Okay, let's get started. Let's do the news. What do we have this week?
This week, we have Lex, Transcribe, Textract, and Personalize. Just high-level services this week. Let's start with Lex. Lex is our chatbot service. This is actually a big deal. You can now save conversation logs for Lex, whether you're interacting with the chatbot in text mode or in voice mode. So you can have text transcripts and audio transcripts. This is something people have been asking for since the service came out. I can hear a lot of people rejoicing in the background. Yes, it's here.
How do you set it up? It's super easy. You just go to the Lex console, and in the settings for your chatbot, you'll see a new entry called conversation logs. This is where you tick the box that says text logs or audio logs or both. You can pass a CloudWatch log group name for text logs as well as an S3 bucket name for audio logs. Don't forget to set the IAM role that gives Lex permission to write to CloudWatch logs and S3. That's all it takes. Another benefit is you'll be able to see missed utterances, sentences that the user said that didn't trigger anything in the bot. Now, it's all logged, so you can run all kinds of analytics. Again, I hear you rejoicing and enjoy this new feature.
Let's move on to Textract. Textract is our text extraction service from images and documents. It's now PCI DSS certified, so if you're building retail and e-commerce applications, Textract is in the picture. More importantly, we released a bunch of quality improvements for Textract. It's even better now at extracting data from tables and forms, which can be quite complicated. I won't spend too much time here because I actually recorded a video demo. I'll put the link to the YouTube video in the description, showing Textract with different types of documents, and it did do pretty well. If you've never tried Textract, this could be a good time to do it.
Let's move on to Personalize, one of my favorite services. Personalize lets you build recommendation and personalization models from CSV data sets. Now, in those data sets, you can pass contextual information, such as device information, time of day, etc. This is important to provide the best possible recommendations. You're probably not looking at the same content on your mobile phone and on the web. So, in the data set you pass to Personalize, you can now use new keywords like location or device to pass that extra information. Some of the algorithms available in Personalize will actually use that data. This allows you to improve the quality of recommendations and build better models. This is pretty cool and easy to try, just use those new keywords.
Amazon Transcribe was also updated this week with two new features. The first one is job queuing, which lets you submit up to 10,000 jobs concurrently. Call centers are heavy users of Transcribe and could have thousands of calls they want to transcribe per day. The prior limit was 100, so you could only submit 100 jobs and then wait for them to complete before submitting others. That's a low number for a call center. Now, we've bumped this number to 10,000, and we have a FIFO queue to process them. Pretty useful if you want to transcribe at scale.
The second feature that Amazon Transcribe received this week is vocabulary filtering. The purpose here is to remove unwanted words from the transcription outputs. You list those words in a text file, one word per line, and upload that file to Transcribe. I built a file containing profanity because that's my use case here. Once you've done that, you can create a transcription job with that filter. Let's create a job and give it a name. The language is English. I'll pass the location of that file in S3. Then I just click Next and say, "Please use my profanity filter." I've got two options: either to completely remove those words or replace them with a triple star token. I'll go for the token. Then I click on create. Let's wait for a minute for this job to complete, and then hopefully, we can see a clean and family-friendly output.
After a couple of minutes, the job is complete. If I open it, look at the transcription. All those nasty words have been removed. "I'm calling because I bought this piece of *** product from your company, and it's complete ***. The *** who sold this to me just told me a whole lot of ***. I just want my money back." So, that's the funniest feature I've tested in a while, and I'm sure it's going to be a popular one.
Now, let's talk about a few extra resources that you may like. The first one is a multi-part video where I'm showing you an end-to-end demo of annotating images. I'm using semantic segmentation because we have a new automatic tool for this. It's a pretty cool demo. If you're interested in labeling image data sets, you're going to like that.
The second thing I want to talk about is my Textract demo. I mentioned earlier we improved the quality of Textract for tables and forms. I recorded a video showing you how to do this.
The last one I want to mention is my DeepSeek library video, introducing you to graph convolutional networks. This one is definitely more technical and code-level, but if you're curious about graph networks, this is a good place to start. You can run the code yourself.
Well, this is it for this first episode, going down in history, maybe or maybe not. We'll see. I hope this was fun. I would love to hear your feedback. Again, if you're building cool stuff on AWS, please get in touch and you can be my guest on the podcast. I'll see you later and keep rocking!