Using artificial intelligence to automate content moderation and compliance with AWS services
Using artificial intelligence to automate content moderation and compliance with AWS services
Customers are learning that content moderation processes by humans alone cannot scale to meet safety, regulatory, and operational needs in the user-generated content era. Artificial intelligence can help gaming, social media, e-commerce, and advertising organizations moderate the deluge of content to reclaim up to 95% of the time their teams spend moderating content manually. Financial, healthcare, and education organizations can streamline the detection and protection of personally identifiable information (PII) across environments and processes.
Learning Objectives:
* Objective 1: Scale to moderate high volumes of User Generated Content (UGC) efficiently.
* Objective 2: Save content moderation time and costs.
* Objective 3: Get started to streamline your content moderation processes.
***To learn more about the services featured in this talk, please visit: https://aws.amazon.com/machine-learni… Subscribe to AWS Online Tech Talks On AWS:
https://www.youtube.com/@AWSOnlineTec…
Follow Amazon Web Services:
Official Website: https://aws.amazon.com/what-is-aws
Twitch: https://twitch.tv/aws
Twitter: https://twitter.com/awsdevelopers
Facebook: https://facebook.com/amazonwebservices
Instagram: https://instagram.com/amazonwebservices
☁️ AWS Online Tech Talks cover a wide range of topics and expertise levels through technical deep dives, demos, customer examples, and live Q\u0026A with AWS experts. Builders can choose from bite-sized 15-minute sessions, insightful fireside chats, immersive virtual workshops, interactive office hours, or watch on-demand tech talks at your own pace. Join us to fuel your learning journey with AWS.
#AWS
Content
2.14 -> [Music]
7.279 -> hi my name is john rouse i'm the
9.2 -> worldwide business development manager
11.36 -> for aws and joining me for today's
14.16 -> webinar using ai to automate content
16.96 -> moderation and compliance with aws
19.68 -> services is nate bochmeier senior
22.16 -> solution architect from aws
24.56 -> as well as ori milinova director of ai
27.119 -> and data science software
30 -> today we're going to talk about creating
31.679 -> safe online environments protect your
33.76 -> brand and minimize moderation costs
37.36 -> first we're going to take a look at user
38.96 -> generated content
41.44 -> then nate is going to show
43.44 -> how to bring in aws ai services together
46.8 -> with a demo
48.879 -> yuri our partner from software is going
50.559 -> to come and talk about how they can help
52.399 -> customers that have content moderation
54.399 -> needs
55.6 -> and then
56.64 -> we'll explain how you get started what
58.32 -> are the next steps
60.079 -> during this session please feel free to
62 -> send any questions during the
63.6 -> presentation as we have experts in the
65.6 -> background ready to help so let's jump
68.159 -> in
70.88 -> modern web and mobile platforms fuel
73.68 -> businesses and drive user engagement
75.92 -> through social features
78.32 -> the daily volume of user-generated
80.32 -> content ugc
82.799 -> and third-party content has been
84.799 -> increasing substantially in industries
88.08 -> such as social media
90 -> social gaming
91.68 -> online forums dating matrimonial and
95.52 -> photo sharing websites
98.479 -> in turn customers need to review audio
101.759 -> image video and text content to ensure
105.92 -> that their end users are not exposed to
107.84 -> potentially inappropriate or offensive
110.479 -> material
111.68 -> such as profanity
113.36 -> violence
114.64 -> drug use adult products nudity or
117.68 -> disturbing content
120.399 -> more than 86 percent of companies today
123.68 -> use ugc as part of their marketing
126.88 -> strategy
132.16 -> this diagram highlights that most modern
135.2 -> businesses are heavily dependent on ugc
139.2 -> and that ugc's growth is outpacing human
142.8 -> capacity
144 -> making content moderation inefficient
146.879 -> unnecessary
148.319 -> risky and expensive
151.92 -> the ugc platform industry is growing at
154.72 -> 26 percent
156.56 -> compounded annually growth rate
158.879 -> and is expected to reach 10 billion
161.84 -> by 2028.
164.239 -> 79 of consumers purchase decisions are
167.12 -> influenced by ugc
169.28 -> and 80 of all web content is ugc
173.519 -> from images on social media to reviews
175.84 -> of products consumers are dominating the
178.56 -> online space
184.4 -> from startups to enterprise customers
187.12 -> need to ensure that their end users are
189.599 -> not exposed to potentially inappropriate
192.56 -> or offensive material or disturbing
194.48 -> content
196.08 -> content moderation is fundamental in
198.56 -> protecting online communities their
201.04 -> members and members personal information
205.04 -> there are strong business reasons to
206.879 -> reconsider how your organization
209.68 -> moderates content
217.36 -> online community members expect safe and
220.4 -> inclu inclusive
222.319 -> experiences where they can freely
224.319 -> consume and contribute images video text
227.84 -> and audio
229.2 -> the ever increasing volume
231.36 -> variety and complexity of ugc
234.64 -> makes traditional human moderation
236.56 -> workflows challenging to scale to
238.959 -> protect users
241.92 -> these limitations force customers into
245.12 -> inefficient expensive
247.439 -> reactive
248.56 -> mitigation processes
250.72 -> that carry an unnecessary risk of users
254.4 -> and the business
256.479 -> the result is a poor harmful and
259.519 -> non-inclusive community experience
262.32 -> that disengages users
264.4 -> negatively impacts community
266.72 -> and business objectives
270.16 -> some statistics
271.6 -> about user generated content
273.919 -> over fifty percent of people said they
276.16 -> create content at least once daily
279.04 -> with twenty three percent saying that
280.8 -> they create more frequently two to five
283.36 -> times per day
285.44 -> more than 40 40 percent of respondents
288.24 -> will disengage from a brand's community
290.8 -> after as little as one exposure to toxic
294.08 -> or fake ugc
296.4 -> while 45 say they will lose all trust in
299.6 -> a brand
301.919 -> 70 of survey respondents stated that
304.4 -> brands need to protect users from toxic
306.8 -> content and 78
309.12 -> said that it's a brand's responsibility
311.44 -> to provide
312.72 -> positive and welcoming online experience
322.24 -> what are customers doing to cope with
324.08 -> cost volume and complexity tied to
327.28 -> moderation
328.88 -> they're turning to ai
330.88 -> ml and other technologies such as deep
333.84 -> learning dl
335.52 -> and natural language processing nlp
338.8 -> to cope with the increase in volume
341.36 -> complexity and the cost to moderate
344.24 -> content accurately and efficiently
349.039 -> customers in modern industries such as
351.199 -> social media fantasy sports and gaming
354 -> and others on traditional verticals
356.88 -> such as financial services and health
358.88 -> care
360.56 -> and facing the same needs and issues
363.44 -> the goal is to improve safe online
365.84 -> environment and reduce moderation costs
376.08 -> gone are the days where consumers
378 -> blindly accept content the way it's been
380.319 -> traditionally served
382.56 -> customers need to ensure that their end
385.28 -> users are not exposed to potentially
388.24 -> inappropriate
389.6 -> or offensive material or disturbing
392.08 -> content
394.08 -> the solution is
395.759 -> scalable content moderation workflows
399.52 -> that rely on artificial intelligence ai
402.96 -> and machine learning ml and deep
404.96 -> learning dl
406.639 -> in natural language processing and lp
408.88 -> technology
412.08 -> these constructs translate transcribe
414.88 -> recognize detect mask redact and
418.56 -> strategically bring human talent into
420.96 -> moderation workflow
423.36 -> to run the actions needed to keep users
425.919 -> safe and engage while increasing
428.08 -> accuracy
429.36 -> and process efficiently in lowering
432.16 -> operational costs
439.05 -> [Music]
440.8 -> aws content moderation ai services can
443.919 -> be leveraged to streamline and automate
447.52 -> your moderation workflows and lower
450 -> operational costs
451.599 -> in the process
454 -> customers can fully manage image video
457.28 -> text
458.319 -> and speech moderation apis
461.199 -> to proact to proactively detect
463.84 -> inappropriate
465.36 -> unwanted or offensive content at scale
468.96 -> and increase brand safety for you and
470.879 -> your partners
473.68 -> image moderation you can detect explicit
475.919 -> adult suggestive content in both image
478.479 -> and videos
480.08 -> video moderation labels are organized in
482.56 -> a hierarchical taxonomy
484.879 -> that provides both top level categories
487.44 -> such as suggestive
489.12 -> and nuanced second
490.96 -> level categories that identify the
493.759 -> specific types of content such as female
496.8 -> swimwear
498.56 -> or partial nudity using this information
501.36 -> you can create granular business rules
503.599 -> for different geographies
505.52 -> target audiences time of day and so on
509.759 -> text detection can be used for image and
511.84 -> video to read text and then check it
513.839 -> against your own list
515.519 -> of prohibited words or phrases
518.56 -> if you want to further analyze text you
521.279 -> can use nlp
525.2 -> audio moderation allows you to detect
527.279 -> profanities or hate speech in videos you
530.16 -> can convert speech to text and then
532.64 -> check it against similar lists
540.16 -> startups social media gaming
543.12 -> and other industries must ensure their
545.44 -> customers collaborate responsibly while
548.48 -> keeping operational costs down
551.2 -> businesses in the broadcasting and media
553.519 -> industries often find it difficult to
556.08 -> efficiently add ratings to content
558.24 -> pieces and formats to comply
561.04 -> with guidelines for different markets
563.04 -> and audiences
564.72 -> other organizations in financial and
566.959 -> health care services
569.2 -> find it challenging to protect personal
571.12 -> identifiable and health information pii
574.32 -> and phi across internal and external
577.44 -> environments and processes
580.24 -> addressing your content moderation needs
582.08 -> require a combination of computer vision
584.56 -> cv and text and language
587.279 -> transform and other ai and ml
589.36 -> capabilities to efficiently moderate the
592.48 -> increasing flux of user-generated
594.88 -> content and sensitive information
598.64 -> aw aws is ai services for moderation and
602.56 -> contextual insight and human in the loop
605.44 -> moderation
606.72 -> can be leveraged to streamline and
608.88 -> automate your image and video moderation
611.2 -> workflows
613.04 -> moderation across one two or more media
616.48 -> types
617.519 -> and data sources plus the addition of ai
620.56 -> insights such as sentiment
622.959 -> contacts and human reviews to either
626 -> guarantee
627.44 -> very high levels of accuracy and to
630.32 -> continue polishing your prediction
632.48 -> models
633.519 -> gives you a complete solution to solve
636.16 -> the most pressing
637.68 -> needs of the fastest growing ugc and
640.56 -> compliance use cases
643.839 -> adding ai insights and connections
646 -> moderation workflows with additional
648.399 -> contextual information
650.32 -> human reviews
651.92 -> smaller sets of moderators to correct
654.56 -> predictions or comply with specific
656.959 -> businesses
658.48 -> so let's start with the social media use
661.36 -> case
662.91 -> [Music]
666.24 -> social media with content moderation
668.24 -> prevents your user exposure to
670.079 -> inappropriate content on photo and video
672.8 -> sharing platforms
674.48 -> such as gaming communities and data
676.88 -> dating applications
678.8 -> these protections increase community
681.04 -> growth
682.399 -> session length conversion metrics and
685.36 -> other responsible social media
687.68 -> objectives and network matrix
691.92 -> this ensures that the content is not
694 -> just appropriate
695.519 -> it also
697.12 -> protects your audience from possible
699.12 -> bullying
700.079 -> or trolling by some
702.399 -> irrational users
704.399 -> but also aligns with your branding and
706.72 -> helps you achieve your overall business
709.92 -> goals
715.36 -> coffee meets bagel is a dating
717.519 -> application that serves potential
719.279 -> matches to over 1.5 million users daily
723.12 -> their motto is quality over quantity
726 -> because they focus on bringing a fun
728.32 -> safe and quality dating experience that
730.959 -> results in meaningful relationships
733.68 -> to deliver on these promises every match
735.92 -> they serve has to fulfill a strict set
738.72 -> of criterias that their users request
743.04 -> coffee meets bagels solution identifies
745.76 -> as user-generated photos that need
749.04 -> moderation and automatically reviews
751.44 -> photo libraries of any scale to detect
754.32 -> unsafe content and apply rules to meet
757.44 -> geographic requirements
760.639 -> they were able to lower moderation costs
762.959 -> by 72 percent
764.8 -> through 97 percent less human
767.12 -> involvement in moderation processes
769.92 -> and lower time for photo approvals from
772.8 -> hours to minutes
777.92 -> like with most communities
780.72 -> gaming companies are developing positive
783.04 -> play guidelines to help make sure your
785.76 -> games and services
788.079 -> are an enjoyable experience for all
790.48 -> players
792.16 -> prevent inappropriate content such as
794.24 -> hate speech profanity or bullying within
797.76 -> a game chat
799.2 -> additionally
800.56 -> moderating user generated values such as
803.2 -> nicknames and profiles
805.04 -> keeps gamers engaged and active and
807.12 -> without motive to leave the game's
809.44 -> ecosystem
812 -> whether you're new to gaming or have
813.76 -> been an active player for years
815.92 -> gaming companies need your help to make
818.24 -> this a community where we all want to be
821.04 -> part of
828.24 -> social is a leading mobile software
830.56 -> company focused on building social
833.279 -> networking and gaming apps
835.36 -> the company develops
837.6 -> omelette arcade a global community where
840.639 -> tens of millions of mobile gaming live
843.36 -> streamers and esport players gather to
846.32 -> share
847.199 -> game play and meet new friends
851.279 -> mobi social solution uses amazon
853.839 -> recognition moderation api
857.6 -> and they use a two-level hierarchical
860.399 -> taxonomy to label categories of
862.72 -> inappropriate or offensive content
866.88 -> they were able to reduce manual content
869.519 -> moderation by 95 percent
872.399 -> increase accuracy and scalability of
874.639 -> operations
875.92 -> and enabling their engineering resources
877.839 -> to focus on
879.68 -> core business areas
887.04 -> content moderation is a process of
888.8 -> monitoring user generated content such
891.279 -> as ecommerce sites
893.76 -> keep out illegal or controversial
896.24 -> product listings that violate compliance
898.959 -> policies that could incur both liability
902.32 -> and buyers and sellers churn
905.199 -> it is the appropriate way to safeguard
907.519 -> the brand's image and manage unwanted
910.24 -> content
912.079 -> in simple terms it's the process of
914.079 -> reviewing
915.199 -> filtering the social media content which
918.079 -> is in form
919.839 -> of con comments images and etc
923.68 -> so you ingest product descriptions and
925.68 -> listings from multiple markets you
927.6 -> detect content types you moderate
930.24 -> through pre-trained or custom models
932.8 -> examples would be in the real estate
934.88 -> auto or auction platforms
938.48 -> within content moderation the values are
940.56 -> it improves search engines ratings
943.759 -> proper content moderation improves your
945.6 -> search rankings organically and helps
947.6 -> businesses to flourish and get a better
950.16 -> online presence
952.48 -> increase brand reputation
954.8 -> as well as help to understand target
956.959 -> audience and scale promotion campaigns
963.839 -> 11th street is an online shopping
966.399 -> company
967.519 -> they're using content moderation to
969.839 -> automate the review of images and videos
973.36 -> as part of 11th street's interactive
975.759 -> experience and to empower their
977.839 -> community to express themselves
980.639 -> they have a feature where users can
982.639 -> submit a photo or video review of the
985.44 -> product that they have just purchased
987.92 -> example wearing
989.519 -> the new makeup
991.92 -> to make sure that no images or videos
993.839 -> contain content that is prohibited by
996.24 -> their platform guidelines they
998.16 -> originally resorted to a manual content
1000.24 -> moderation
1001.44 -> they quickly found that this was costly
1003.839 -> error prone and not scalable
1006.56 -> eleventh street solution now uses
1009.279 -> amazon recognition moderation labels and
1012.399 -> video apis
1014.72 -> to automate moderation of thousands of
1017.12 -> customer photo and video
1019.519 -> reviews daily
1022 -> 11th street now reviews more than 7 000
1025.12 -> images and videos every day
1027.52 -> with higher quality
1029.28 -> speed and at a lower cost when compared
1032.079 -> to its initial manual moderation
1034.319 -> approach
1039.28 -> within advertising
1042 -> content moderation helps avoid
1044.079 -> associations that increases the risk of
1047.12 -> public backlash due to unwanted
1049.84 -> association between your brand
1052.4 -> and ad
1053.44 -> or content within ads
1056.24 -> brand safety has become a major issue in
1058.96 -> the online media
1060.799 -> industry
1061.919 -> worldwide
1064.16 -> companies who value their brand should
1066.24 -> be concerned about what web pages their
1068.799 -> ads appear on
1070.32 -> and then next
1071.919 -> to what kind of content many advertisers
1074.64 -> assume that their brands are safe if
1076.48 -> they're running their online campaigns
1078.799 -> with a major media agency
1080.96 -> a major media exchange or platform
1084.32 -> or major network
1086.96 -> in fact brand safety should never be
1088.799 -> assumed many major media agencies place
1092.32 -> their advertiser ads and very bad media
1095.28 -> placement that can harm brand image and
1098.08 -> in some cases can be potentially lead to
1100.799 -> legal action
1107.76 -> flipboard is one of the world's first
1110.559 -> social magazines
1112.559 -> inspired by the beauty and ease of print
1115.52 -> media
1116.559 -> the company's mission is to
1118 -> fundamentally improve how people
1120.24 -> discover view and share content across
1123.36 -> their social networks
1126.16 -> flipboard is a content recommendation
1128.72 -> platform that enables publisher
1131.44 -> creators and curators to share stories
1134.32 -> with readers to help them stay up to
1136.32 -> date on their passions and interests
1139.679 -> on average flipboard processes
1141.52 -> approximately 90 million images per day
1144.799 -> to maintain a safe and inclusive
1146.72 -> environment and to confirm that all
1148.96 -> images comply with platform guidelines
1151.28 -> at scale
1152.72 -> it is crucial to implement a content
1155.12 -> moderation workflow using machine
1157.2 -> learning
1159.2 -> however building models for this
1162.08 -> for this system internally was labor
1164.88 -> intensive and lack the accuracy
1166.72 -> necessary to meet the high quality
1168.72 -> standards
1169.84 -> flipboard's users expect
1172.4 -> this is where amazon recognition became
1175.039 -> the right solution
1176.72 -> for their product
1183.84 -> for the highly regulated industries
1186.48 -> financial services there's a need to
1188.799 -> detect
1189.919 -> and redact
1191.28 -> pii to ensure that sensitive user data
1194.96 -> remains private
1196.64 -> your customers can trust your platform
1199.039 -> and increase participation
1201.36 -> investment and referrals
1204.72 -> healthcare there is a need to detect and
1207.039 -> redact pi phi
1209.679 -> and other sensitive information to
1211.44 -> ensure that data remains private
1214.159 -> healthcare providers can remain
1216.24 -> compliant with hipaa and other
1218.799 -> regulatories to avoid fines
1222.24 -> legal brief management
1224.08 -> automate extraction of insight from
1226.24 -> packets on legal briefs such as
1228.64 -> contracts and court records
1231.919 -> further secure your documents by
1233.679 -> identifying and redacting personally
1236.08 -> identify a personally identifiable
1239.679 -> information pii
1242.4 -> process financial documents classify and
1245.2 -> extract entities
1246.799 -> from financial services
1248.799 -> documents such as insurance claims or
1251.28 -> mortgage packages or fine relationships
1254.4 -> between financial events and a financial
1257.84 -> article
1259.36 -> and an example would be to analyze
1261.52 -> support tickets and knowledge articles
1264 -> to detect
1265.44 -> pii entities and redact the text from
1268.88 -> your index
1270.72 -> the documents and search solutions after
1273.36 -> that search solutions are free of pii
1275.919 -> entities and documents
1277.76 -> redacting pii entities help you protect
1280.88 -> privacy and comply with the local laws
1283.6 -> and regulations
1293.919 -> pterodact solutions
1295.919 -> software offers a robust alternative to
1298.64 -> secure information
1300.4 -> sharing in a world of ever-increasing
1302.84 -> compliance and privacy concerns
1306 -> with its signature information
1309.679 -> and presentations
1311.44 -> and capabilities
1313.039 -> pteradac's tools provide the users with
1315.2 -> a safe
1316.559 -> information and sharing
1319.039 -> environment
1326.72 -> addressing your content moderation needs
1328.799 -> requires a combination of computer
1330.88 -> vision text and language
1333.44 -> transform and other aiml capabilities to
1336.559 -> efficiently moderate the increasing
1339.12 -> influx of user-generated content and
1341.6 -> sensitive information
1343.76 -> aws is ai services amazon recognition
1347.52 -> amazon transcribe
1349.36 -> amazon comprehends and amazon translate
1353.039 -> can be leveraged to streamline and
1354.799 -> automate your image and video moderation
1356.72 -> workflows
1358.96 -> i'd like to hand it over to nate to talk
1360.88 -> more about these services and solutions
1364.88 -> let's examine how you can implement low
1366.96 -> code workflows using aws ai services and
1370.48 -> serverless technologies
1375.12 -> you'll need scalable content moderation
1377.039 -> workflows that rely on artificial
1378.88 -> intelligence machine learning deep
1381.12 -> learning and natural language processing
1383.12 -> technologies
1384.32 -> these technologies must support
1385.919 -> translation transcription recognition
1388.88 -> entity detection pii masking compliance
1392.32 -> redaction and strategically bring human
1394.72 -> talent into the moderation workflow
1397.52 -> you run the actions as needed to keep
1399.52 -> your users safe and engaged while
1401.6 -> increasing accuracy process efficiency
1404.32 -> and lowering operational costs
1408.48 -> you don't need expertise in ml to
1410.559 -> implement these workflows and can tailor
1412.559 -> these patterns to your specific business
1414.24 -> needs aws delivers these capabilities
1417.2 -> through fully managed services that
1419.12 -> remove complexity and undifferentiated
1421.44 -> heavy lifting without requiring a data
1423.6 -> science team
1427.039 -> let's review the five core aws ai
1429.679 -> services that you'll need and how they
1431.84 -> integrate into a holistic solution
1436.48 -> first amazon recognition makes adding
1439.44 -> photo image and video analysis to your
1441.919 -> application easy using its api the
1444.32 -> service can identify objects people
1446.88 -> texts scenes and activities out of the
1449.52 -> box it can also detect inappropriate
1451.76 -> content and return hierarchical
1454.08 -> categorical labels like drugs and
1455.84 -> alcohol
1457.36 -> amazon recognition also provides highly
1459.84 -> accurate facial analysis face
1462.32 -> comparisons and face search capabilities
1464.96 -> for instance liquor establishments do
1467.12 -> not want images of underage drinking on
1469.2 -> their forums
1472.799 -> next amazon transcribed is an automatic
1475.279 -> speech recognition service that uses ml
1477.76 -> models to convert audio to text
1480.72 -> with amazon transcribed you can ingest
1483.12 -> audio files and real-time audio streams
1486.08 -> the service produces easy to read
1487.919 -> transcripts improves accuracy with
1490.32 -> language customizations and filters
1492.64 -> content to ensure customer privacy
1495.52 -> after converting the audio to text you
1497.52 -> can analyze it further with other aws
1499.679 -> services
1503.12 -> third amazon translate is a text
1505.279 -> translation service that uses advanced
1507.52 -> ml technologies to provide high quality
1510.08 -> translations on demand
1512.88 -> global customers expect to collaborate
1514.48 -> with social platforms in their native
1516.08 -> language meeting this expectation would
1518.08 -> be complex except that you can use
1520 -> amazon translate to convert text to over
1522.88 -> 70 languages and variants in over 15
1525.44 -> regions it also supports profanity
1527.76 -> masking during the translation without
1529.6 -> requiring custom code
1534.08 -> fourth amazon comprehend uses natural
1536.48 -> language processing to extract insights
1539.12 -> about the contents of a document or text
1542.08 -> it develops insights by recognizing the
1544.08 -> entities the key phrases the language
1546.159 -> and sentiment among other common
1548.08 -> elements in a document with comprehend
1551.039 -> your code focuses on using nlp results
1553.919 -> not parsing text
1556.799 -> you can run a real-time analysis for
1559.44 -> small workloads or start asynchronous
1561.76 -> analysis jobs for large document sets
1565.76 -> you can use the pre-trained models that
1567.84 -> amazon comprehend provides or train your
1570.24 -> custom models for classification and
1572.159 -> entity recognition
1576.159 -> finally amazon augmented ai makes
1578.64 -> building and managing human reviews for
1580.799 -> content moderation and ml workflows easy
1584.4 -> you can specify that human reviewers
1586.24 -> only step in when a model cannot make a
1588.72 -> high confidence prediction
1590.559 -> or to audit the model on an ongoing
1592.799 -> basis
1599.36 -> you can build content moderation
1600.88 -> workflows using cloud first serverless
1603.2 -> components with low code implementations
1605.919 -> this approach removes the challenges of
1607.52 -> managing and scaling infrastructure and
1609.36 -> allows your devops team to focus on
1611.36 -> delighting your customers
1613.76 -> here you can see the standard reference
1615.44 -> architecture for implementing content
1617.279 -> moderation workflows for web and mobile
1619.679 -> applications
1621.84 -> first your solution ingests user
1623.84 -> generated content through its front-end
1625.84 -> interface users expect this experience
1628.159 -> to be interactive and engaging
1630.799 -> aws amplify is a set of purpose-built
1633.52 -> tools and features for building those
1635.679 -> full stack experiences and integrating
1637.679 -> into aws's breadth and depth
1642.08 -> typical integrations include
1643.44 -> capabilities like api management and
1645.84 -> content delivery networks
1647.76 -> you can meet both requirements securely
1650.24 -> economically and at scale using services
1653.2 -> like amazon api gateway and amazon
1655.919 -> cloudfront
1659.84 -> second you need a place to stage that
1662 -> incoming content and amazon s3 comes
1664.48 -> into play here s3 is our object storage
1667.44 -> service
1668.799 -> that can persist your images audio video
1671.6 -> and rich text documents at virtually any
1673.52 -> scale
1674.88 -> the five previously discussed aws ai
1677.44 -> services directly support reading from
1679.679 -> an s3 bucket
1683.84 -> third after staging the content you can
1686.08 -> trigger asynchronous event driven
1687.84 -> processing with services like amazon
1690.159 -> eventbridge eventbridge is a serverless
1692.559 -> event bus with advanced publisher
1694.48 -> subscriber features for using events
1697.039 -> generated from your application and aws
1699.679 -> services
1701.12 -> another powerful pattern is to combine
1704.24 -> amazon api gateways service integrations
1707.279 -> with amazon step
1708.84 -> functions this combination lets you
1711.12 -> elastically execute state machines that
1713.919 -> reliably orchestrate distributed
1715.76 -> applications
1717.6 -> last year we announced that step
1718.799 -> functions can now directly call over
1720.88 -> 9000 aws sdk apis without any additional
1726.32 -> code
1728.32 -> i recently published a blog containing
1730.08 -> prescriptive guidance for the next three
1731.919 -> steps stick around until the end of the
1733.919 -> presentation
1735.2 -> for the link
1737.36 -> fourth
1738.72 -> moderating images requires detecting
1740.64 -> inappropriate unwanted or offensive
1742.799 -> content containing suggestiveness
1745.2 -> violence and other categories from
1746.72 -> pictures
1748.32 -> videos can reuse image moderation
1750.48 -> processes using a frame sampling
1752.72 -> procedure
1758.48 -> fourth moderating images requires
1760.48 -> detecting inappropriate unwanted or
1762.559 -> offensive content containing
1764.32 -> suggestiveness violence and other
1766.559 -> categories from pictures
1768.88 -> videos can reuse image moderation
1771.279 -> processes
1772.72 -> by implementing a frame sampling
1774.64 -> procedure
1776.399 -> after selecting the image or frame use
1778.96 -> amazon recognitions built-in apis
1782.08 -> to detect inappropriate content
1784.32 -> and read text that may be within the
1786.399 -> image
1787.52 -> the apis return bounding boxes for each
1789.919 -> entity found reporting why it was
1791.84 -> flagged
1793.12 -> amazon recognition also supports text
1795.2 -> detection written in english and seven
1797.36 -> other languages
1799.44 -> in addition amazon recognition can
1801.44 -> quickly review millions of images or
1803.84 -> thousands of videos using ml and flag
1806.399 -> the subset of assets requiring further
1808.799 -> action
1809.84 -> pre-filtering helps provide
1811.679 -> comprehensive yet cost efficient
1813.76 -> moderation coverage while reducing the
1816.159 -> content that human teams have to review
1822.08 -> to moderate audio files you must
1824.32 -> transcribe the file to text and then
1826.559 -> analyze it this process has two variants
1829.039 -> depending on whether you're processing
1830.72 -> individual files synchronously
1833.039 -> or live audio streams asynchronously
1835.919 -> synchronous workflows are ideal for
1837.76 -> batch processing where the caller
1839.6 -> receives one complete response in
1841.76 -> contrast audio streams require periodic
1844.399 -> sampling with multiple transcription
1846.72 -> results
1848.159 -> amazon transcribe is an is an automatic
1851.2 -> speech recognition service that
1852.88 -> integrates into synchronous workflows
1855.679 -> by starting the transcription job and
1857.76 -> then periodically polling
1859.76 -> for the jobs completion
1862.32 -> live audio streams need a different
1864.08 -> pattern that supports a real-time
1865.6 -> delivery model
1867.679 -> streaming can include pre-recorded media
1870.48 -> such as
1871.84 -> movies music podcasts
1874.24 -> or real-time media such as live news
1876.96 -> broadcasts you can transcribe audio
1879.76 -> chunks instantaneously using the
1881.44 -> streaming api
1883.44 -> over http 2
1885.519 -> or websockets protocols
1888 -> after you receive the transcription
1889.44 -> results you can reuse existing plain
1891.919 -> text moderation workflows to assess the
1894.24 -> audio content
1898.559 -> next you can use amazon text direct to
1901.039 -> extract handwritten text and data from
1903.279 -> scanned documents you asynchronously
1905.519 -> begin an analysis report by specifying
1908.88 -> one or more word documents or pdf
1911.12 -> documents
1912.64 -> the analysis will discover pages
1915.279 -> paragraphs tables and key value pairs
1918 -> within the document cohort this
1920.24 -> contextually detailed information can
1922.399 -> power intelligent document processing
1924.24 -> pipelines that moder moderate and redact
1927.519 -> at specific field levels for example a
1930.24 -> health provider might redact patient
1931.919 -> names on every non-signature page
1938.32 -> after running the moderation workflow
1940.399 -> some data will report low confidence
1942.48 -> scores for these situations you can push
1945.12 -> that content into amazon sagemaker
1947.519 -> ground truth and orchestrate data
1949.6 -> labeling jobs
1950.96 -> ground truth supports private human
1953.519 -> workforces and crowdsourcing through
1955.76 -> amazon mechanical turk
1957.679 -> you can import that label data into
1959.6 -> services like amazon recognition and
1961.679 -> comprehend to build custom entity and
1963.919 -> vocabulary detection models
1967.2 -> this capability allows you to easily
1969.6 -> extend the previously discussed
1971.44 -> moderation workflows
1973.279 -> now for many customers they need
1975.44 -> something that's between a private
1976.96 -> workforce and the public amazon
1979.44 -> mechanical turk for these situations you
1982.08 -> can leverage amazon ground truth plus a
1984.399 -> turnkey solution for your data labeling
1986.559 -> needs
1989.84 -> lastly as previously mentioned amazon
1992 -> a2i lets you define policies that bring
1995.44 -> human reviewers into the conversation as
1997.919 -> necessary
2001.76 -> next i'm going to demo the amazon media
2004.24 -> to cloud solution you can download media
2006.48 -> to cloud from the amazon solution
2008.399 -> library and deploy it into your account
2011.6 -> the solution takes about 25 minutes to
2013.679 -> provide a complete website with
2015.12 -> capabilities to experiment with the core
2017.44 -> aws ai services across audio video and
2020.559 -> rich text documents i have already
2022.72 -> preloaded the solution with a few
2024.08 -> documents so now let's see it in action
2026.799 -> today i'm going to be giving a quick
2028.399 -> demo of media to cloud
2031.039 -> you can download this at the aws
2033.679 -> solutions library
2035.519 -> once you're at this page just scroll
2037.12 -> down and click launch in aws console
2040.399 -> follow the guide and it will get this
2042.32 -> installed in your account in about 25
2044.48 -> minutes
2049.52 -> after installing
2051.52 -> this example
2053.2 -> you should be able to log into a
2054.96 -> cloudfront web portal
2057.04 -> from here all you have to do is
2060.48 -> select a handful of example files
2064.399 -> drop them
2065.76 -> over here
2068.159 -> say next
2074.639 -> next
2078.32 -> from here you can take and you can
2080.159 -> control
2081.28 -> all of the different aws
2084 -> ai service parameters for extracting
2087.119 -> content from audio video
2091.28 -> or
2092.56 -> rich text documents
2094.48 -> you can control things like what is the
2096.72 -> confidence
2097.92 -> where you want to cut it off
2100.16 -> you can have custom collections
2102.96 -> and so a collection is a group of
2105.119 -> celebrities that are
2107.28 -> celebrities to you these could be your
2108.96 -> classmates it could be your co-workers
2111.44 -> so on and so forth you can specify that
2114 -> you want to extract text but only from a
2116.56 -> section
2117.839 -> of images
2120 -> if you have custom models let's say for
2122.56 -> example
2123.52 -> that there's something in particular
2125.2 -> that you want to look for
2128.24 -> maybe it's
2129.599 -> uh your logo maybe it's a competitor's
2131.44 -> logo so on and so forth things that you
2133.52 -> just don't want
2135.119 -> side by side with your brand
2137.92 -> any of those types of things you can
2139.599 -> take and define as custom models
2142 -> you can
2142.839 -> specify what are the frame rates that
2145.2 -> you want to sample if it's a video
2148.4 -> how is it you want transcriptions to
2150.079 -> work let's say that the video contains
2152.32 -> audio how do you want that to work
2154.72 -> should it auto detect the language
2156.56 -> should you assume that it's french or
2158.32 -> spanish
2159.44 -> you can also
2160.96 -> provide custom vocabularies and so
2164.4 -> let's say that you wanted to include
2166 -> your product names
2167.839 -> you could specify those as unique
2170.96 -> terms so that they don't get messed up
2173.52 -> during the translation
2175.44 -> similarly for comprehend our natural
2177.68 -> language processing service
2179.599 -> you can specify that you're interested
2181.68 -> in sentiment analysis or key phrase
2184.64 -> detection
2187.2 -> once you've configured this i'm just
2188.64 -> going to leave it with the defaults for
2190 -> right now
2190.96 -> you simply say next and then click start
2194.4 -> i'm going to pause it here it only takes
2196 -> a couple of seconds to uh finish
2198.079 -> importing
2199.28 -> but let's uh see what we get out of this
2203.04 -> so the first example that i want to show
2204.96 -> you
2205.68 -> is a beer video this is a video that i
2208.96 -> shot using my cell phone and a empty
2211.359 -> beer bottle
2214.56 -> so when we click into it
2218.96 -> what we can see is that
2221.359 -> these are all of the different
2222.56 -> operations that ran
2224.96 -> let's say that we're interested in the
2227.04 -> recognition labels let's turn these on
2232.88 -> and hit play
2234.88 -> and so initially
2236.32 -> it says that there's nothing to see here
2238.8 -> but as soon as we reach
2240.8 -> this piece right here
2242.4 -> see well
2243.44 -> using the recognition apis it can
2245.52 -> determine that
2246.8 -> this is where we needed to moderate some
2249.119 -> content we can see that these are
2252.16 -> the types of things that were flagged
2254.8 -> now meanwhile
2256.48 -> as it goes out of frame
2261.04 -> at this point
2262.56 -> so i've turned the bottle backwards
2265.359 -> and it and at this point it says
2267.92 -> that it is not confident enough that
2269.44 -> it's an alcoholic container but it is
2271.359 -> still confident that this is a beverage
2273.119 -> and drink and so this would be a really
2275.04 -> good place for something like augmented
2277.76 -> ai
2279.04 -> and so with a to i
2280.96 -> what you could do is you could say that
2282.72 -> this may be on the edge i don't know if
2285.119 -> we should be moderating this so let's
2287.52 -> actually hand this over to a human to do
2289.599 -> a deeper dive and so once that human
2292 -> receives this video
2293.599 -> oh through a to i they would see
2296.56 -> this is the thing which is being flagged
2299.599 -> as the human looks at it they'll
2301.52 -> probably say yeah that looks like a beer
2303.52 -> bottle
2304.96 -> and whether or not that's appropriate
2306.88 -> depends on
2308 -> your particular situation
2310.64 -> now let's see what happens with
2312.8 -> the photos that we've imported
2315.52 -> first
2316.8 -> we can see that from the moderation
2320.24 -> that it is fairly confident that this is
2322.88 -> some sort of a drug product makes sense
2325.119 -> it's got a marijuana logo on it
2328.88 -> at a minimum it's probably a tobacco
2330.64 -> product
2332.88 -> the other thing that you can see from
2334.64 -> here
2336.4 -> is that
2339.68 -> media to cloud will tell you how long
2341.359 -> each of these operations took
2343.28 -> and so for this particular one
2345.599 -> about
2347.2 -> just a little under a second
2349.28 -> to moderate
2350.88 -> this image which means that this is
2353.04 -> happening fast enough that you could
2354.72 -> have this as part of an interactive
2357.04 -> uh workflow
2358.88 -> now
2359.599 -> let's say that you were going to upload
2360.96 -> some long movie in that case
2363.839 -> it's going to take longer but for simple
2366.079 -> things like images and pdf documents
2369.04 -> things like that you should be looking
2370.88 -> at
2371.92 -> you know
2372.88 -> single-digit seconds even for massive
2375.28 -> documents
2379.04 -> now for the last document that was
2381.28 -> uploaded
2387.599 -> which was a movie poster
2391.92 -> from here
2393.04 -> you can very quickly and easily
2396.32 -> see that using the built-in apis it's
2398.56 -> able to determine that there are four
2400.88 -> people in here
2403.68 -> is able to start
2405.2 -> labeling
2406.4 -> that these are humans
2408.72 -> they're wearing evening
2410.079 -> gowns
2411.2 -> it's fashionable
2412.64 -> and so
2413.68 -> what you can quickly see from this
2416.16 -> and you should try this with your own
2417.52 -> data
2418.48 -> is that you can start extracting out
2421.28 -> things which are relevant and things
2422.88 -> that are not relevant to your particular
2425.04 -> audience
2427.52 -> in this particular case
2429.2 -> it found that these four people are
2430.96 -> media
2432 -> celebrities there's sarah jessica parker
2435.2 -> now like i mentioned earlier you do not
2437.599 -> have to use media celebrities you can
2439.92 -> use custom celebrity groups which
2442.48 -> include your co-workers friends family
2445.44 -> anything along those lines
2447.44 -> now in this particular poster it didn't
2449.359 -> find anything which is which needs to be
2451.839 -> moderated
2455.359 -> but once again
2456.72 -> you know we can come here
2458.96 -> and see that
2460.64 -> you know these operations are not
2462.16 -> exactly
2463.599 -> taking too long they're about a second a
2465.359 -> piece once again
2466.96 -> which means that it's fast enough that
2469.28 -> you can start relying on
2471.68 -> these being part of interactive queries
2476.319 -> thank you nate
2478.64 -> so moderating and innovating with aws
2481.359 -> partners
2483.28 -> we can make content moderation
2485.839 -> easy from free trials to pocs to all the
2489.44 -> way through production
2492.319 -> we recommend that you start with a poc
2494.72 -> to test out the accuracy benefits that
2497.2 -> your business can see
2499.52 -> so contact us to help find the right
2501.44 -> partner or explore the aws partner
2504.4 -> network to get a poc running
2507.2 -> let aws partners like soft serve help
2510.24 -> you turn processes into competitive
2512.48 -> differentiations with unique and cost
2515.68 -> efficient moderation solutions and
2518.56 -> integrations
2520.86 -> [Music]
2525.2 -> soft serve is a technology company
2527.119 -> specializing consulting services and
2529.359 -> software development and an aws partner
2533.44 -> now i'd like to hand it over to yuri who
2536 -> is the director of ai and data science
2538.24 -> for soft surf to talk more about how
2540.8 -> they are working with customers today
2544.64 -> thank you john hello everyone my name is
2547.04 -> yuri malovanov i'm a director of ai in
2549.359 -> data science practice with software a
2551.68 -> premier aws partner specializing in
2554.16 -> applying cutting-edge technologies such
2555.92 -> as ei and machine learning to address
2558 -> complex business challenges across
2559.68 -> various industries including media
2562.319 -> retail finance health care manufacturing
2565.52 -> and energy
2568.16 -> to illustrate how an aws partner like
2571.04 -> sorcerer can help accelerate your ai
2573.76 -> journey let's consider a typical build
2576 -> versus buy trade that many companies
2578.56 -> face when they're looking for a digital
2580.72 -> solution to their business problem
2583.04 -> regardless of the size of your company
2585.52 -> or your particular challenge there are
2587.68 -> always
2588.56 -> plenty of off-the-shelf products that
2590.88 -> aim to address a specific business need
2593.119 -> or a niche on the market although those
2595.68 -> solutions can address many common
2597.52 -> problems out of the box which means fast
2599.76 -> time to market they are often limited in
2602.48 -> customization and integration
2603.92 -> capabilities at the same time the
2606.079 -> availability of new features depends on
2608.4 -> a particular product roadmap
2610.8 -> moreover of the shelf solutions are
2612.8 -> available to everyone which means they
2615.2 -> can provide your business with much
2617.119 -> competitive advantage
2618.72 -> if your challenge goes beyond generic
2620.64 -> capabilities or your problem is very
2623.44 -> specific to your business or if you want
2626.319 -> to differentiate on the market by
2628.079 -> enabling unique to your business
2629.52 -> features and capabilities you should
2631.359 -> instead consider building a solution
2633.2 -> in-house this is where aws partners like
2636.48 -> us can help you innovate and adopt new
2638.88 -> technologies while letting you focus on
2641.28 -> your core business practices and
2642.8 -> objectives in addition to that aws
2645.599 -> provides a powerful cloud infrastructure
2647.76 -> that drastically accelerates ai
2649.52 -> experimentation development and the path
2651.68 -> from idea to production
2654.16 -> the built-in ei services such as amazon
2656.56 -> recognition or comprehend allow you to
2658.72 -> build your solution from building blocks
2661.119 -> and provide you with robust
2662.4 -> customization features a great example
2664.88 -> is the amazon recognition custom labels
2666.96 -> service which simplifies the training of
2669.119 -> custom computer vision models using your
2671.119 -> data
2672.64 -> if you need to go beyond that the aws
2675.119 -> sagemaker platform can streamline your
2677.119 -> custom ml model development process and
2679.52 -> help scale your ai experiments in days
2681.839 -> instead of weeks or even months
2684.4 -> both scenarios allow you to easily
2686.24 -> integrate and augment your ai solution
2688.4 -> with a wide variety of the
2689.76 -> market-leading services provided by aws
2692.24 -> cloud ecosystem
2698.96 -> to give you an example let me tell you
2700.8 -> one of our success stories around
2702.56 -> content moderation in media our customer
2705.44 -> the world's leading toy manufacturer was
2707.76 -> looking for a way to enhance their
2709.599 -> social media platform by automating the
2712.079 -> manual processes of moderating user
2714.16 -> generated content uploaded by the global
2716.72 -> community of their customers who are
2718.8 -> primarily children their initial goal
2721.52 -> was to automate the most of the
2723.04 -> monotonous manual work while keeping the
2725.44 -> more complicated cases for human review
2728.88 -> to operate this challenge first we
2730.88 -> needed to facilitate the business
2732.48 -> requirement collection and frame out an
2734.72 -> ai problem the questions that were
2737.119 -> important at this stage included what
2739.28 -> are the key business drivers and
2740.64 -> priorities behind this problem what type
2743.119 -> of content is considered to be
2744.64 -> inappropriate what are the business
2746.56 -> implications of such content
2748.8 -> what is the potential solution may look
2750.88 -> like
2752.319 -> what data is needed and what data is
2754.24 -> available
2755.28 -> what user personas will interact with
2757.2 -> the solution and what are their pains
2759.28 -> and needs
2760.88 -> finally it's important to rate the
2763.28 -> question of how would we measure the
2764.8 -> success of the solution
2767.119 -> once the problem was framed the team
2769.359 -> could start collecting analyzing the
2771.04 -> data running ml experiments and
2773.359 -> validating early results for the
2775.04 -> business
2776.079 -> thanks to aws the vast majority of the
2778.64 -> challenges were addressed by the
2780.079 -> building aws services and apis that
2783.52 -> allowed us to deploy a pilot solution
2785.44 -> that is ready for production use in a
2787.44 -> matter of weeks
2788.96 -> using amazon recognition we could detect
2791.44 -> human faces on images or real-time video
2794.48 -> streams and differentiate between kits
2796.72 -> and adults pictures
2798.64 -> since amazon recognition doesn't analyze
2801.2 -> the quality of the images our team
2803.28 -> designed and trained a set of custom ml
2805.44 -> models on sagemaker that can identify
2808.079 -> different types of artifacts and defects
2810.16 -> on the photos and provide an overall
2812.24 -> image quality score
2813.92 -> finally amazon augmented ei help us
2816.72 -> incorporate a human in the loop aspect
2818.88 -> to continuously improve the solution
2820.72 -> with new knowledge and user feedback
2822.96 -> while amazon recognition automated the
2824.88 -> filtration of high confidence incidents
2828.079 -> augmented ai allows subject matter
2830 -> expert to analyze edge cases that
2831.839 -> require further evaluation
2833.92 -> all that help our customer achieve
2836.16 -> automated human level of image
2837.68 -> understanding with over 90 of accuracy
2841.04 -> reduce the cost of manual digital
2842.88 -> content moderation and finally uncover
2845.359 -> new insights into the user-generated
2847.599 -> content on their platform
2852.64 -> now let me show you what a typical ai
2854.72 -> customer journey with sorcerer looks
2856.559 -> like we start with the rapidi assessment
2860.24 -> our enablement servers offering that
2862.24 -> helps accelerate ai adoption by
2864.079 -> collaboratively evaluating ideating and
2867.2 -> ultimately aligning your technology and
2869.28 -> business drivers with ai enabled
2870.96 -> solutions it allows identifying and
2873.76 -> prioritizing unique ei opportunities
2876 -> that will further drive success and
2877.92 -> innovation and reduce the risk of your
2880.079 -> eye investments
2884 -> once the challenges are identified our
2886.559 -> in-house design thinking and rapid
2888.319 -> prototyping methodologies allows us to
2891.119 -> drastically reduce the uncertainty in ai
2893.599 -> development and accelerate its time to
2895.52 -> market during this stage our team
2897.92 -> focuses on the most critical components
2900.079 -> and features by validating the key
2902.16 -> hypotheses assumptions and design
2904.4 -> decisions that impact the feasibility
2906.64 -> and the success of your ei solution
2909.76 -> our iterative approach allows to
2911.52 -> gradually increase the complexity and
2913.44 -> fidelity of the solution evolving a
2915.68 -> prototype into a full-fledged production
2917.68 -> ei system
2919.119 -> but before exposing your ai solution to
2921.28 -> the end users it is critical to set up a
2923.76 -> proper framework to measure the success
2926 -> and the business impact of the solution
2928.24 -> our proof of value methodology helps
2930.48 -> address this challenge at the early
2932.319 -> stages by designing a set of
2934.16 -> well-defined and measurable kpis metrics
2937.04 -> and success criteria that resemble your
2939.76 -> business goals and priorities
2942.24 -> while a data science team works on ai
2944.4 -> experiments and solution design our
2946.8 -> envelopes and cloud engineering teams
2948.64 -> can focus on preparing the foundation
2950.8 -> for the future machine learning
2952.24 -> workloads
2953.92 -> including setting up and configuring the
2956.16 -> required cloud infrastructure building
2958.4 -> data by appliance and integrations and
2960.64 -> automating the key machine learning
2962.16 -> operations and workflows that enables
2964.88 -> fast production deployments even after
2966.96 -> first prototyping iterations and
2969.2 -> automates continuous improvements with
2971.119 -> better models and new data when the
2973.04 -> solution is live
2974.88 -> finally we also help our customers
2977.28 -> manage maintain and improve their ai
2979.76 -> solutions in production allowing our
2981.839 -> customers to focus on their core
2983.52 -> business functions
2986.64 -> all right i hope you found this
2988.4 -> information insightful if you are
2990.24 -> interested in learning more about
2991.92 -> sorcerer and what we can do to
2993.92 -> accelerate the innovation within your
2995.52 -> organization feel free to reach out to
2997.68 -> aws or contact me directly thank you
3000.4 -> very much with that back to you john
3005.04 -> yuri thanks so much for sharing on how
3007.76 -> you can help our customers that have
3010.64 -> content moderation
3012.4 -> needs
3013.599 -> so how do you get started
3015.839 -> try all of our moderation ai services
3019.68 -> at no cost
3021.2 -> and access developer guides tutorials
3024.319 -> github repositories use cases
3028.16 -> webinars
3029.44 -> getting started guides pricing and more
3032.24 -> for each of those services
3035.2 -> aws experts
3037.119 -> and solution architects can help you
3038.96 -> with additional information or resources
3041.76 -> needed to get you started
3043.76 -> as mentioned before contact us to help
3046.319 -> you find the right partner like soft
3048.16 -> serve or explore the aws partner network
3051.28 -> to get a poc running today
3058.079 -> yuri thank you so much for sharing on
3060.16 -> how you can help our customers with
3062.64 -> their content moderation needs
3065.44 -> again thank you nate
3067.44 -> and yuri for joining me today
3069.839 -> and please feel free to reach out to us
3072.079 -> if you have any additional questions
3074.319 -> or need help with your content
3077.119 -> moderation needs either via email
3080.559 -> or linkedin also you can find more
3083.04 -> information
3084.559 -> on this specific use case content
3086.48 -> moderation within aws.amazon.com
3090.96 -> machine learning under ml use cases
3098.87 -> [Music]
3104.48 -> you
Source: https://www.youtube.com/watch?v=yMN3Xx0DcoU