Videos o6U_2vd967Y

Let's integrate AI Agents in Event-Sourced Systems — Divakar Kumar, FlyersSoft

index_state ready data_status ok

AI Engineer· published 2026-07-30· 0:21:36· en-US· indexed 2026-08-10 19:38

Open on YouTube

57 shot(s).

keyframes kept every frame deduplicated

What was stored

cues: 204; whisperx 204
chunks: 37; from 204 cues
keyframes: 35; kept of 57 captured
frames with text: 35; 569 lines read
chapters: 11; from the source metadata
keyframe bytes: 6.7 MB; word timings on 204 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage	state	model	started	took
`fetch`	done	—	2026-08-09 21:58	0s
`stt`	done	—	2026-08-09 03:31	21s
`chunk`	done	—	2026-08-09 03:31	0s
`text_embed`	done	—	2026-08-10 19:38	0s
`keyframe`	done	—	2026-08-09 03:31	2m 44s
`ocr`	done	—	2026-08-09 03:34	10s
`frame_embed`	done	—	2026-08-10 19:38	6s

Frames, and what the machine read

0:02 #0 done2 line(s)

shot 0·sharpness 459.9
1. AlEngineer0.96
2. World's Fair1.00
0:03 #1 done2 line(s)

shot 1·sharpness 662.1
1. AlEngineer0.95
2. World's Fair0.99
0:10 #2 done24 line(s)

shot 2·sharpness 2744.8
1. LAB & PLATINUM SPONSORS0.99
2. Amazon AGI Lab0.98
3. ANTHROP\C1.00
4. Google DeepMind1.00
5. MINIMAX0.94
6. OpenAI0.92
7. Akamai1.00
8. arize1.00
9. aws1.00
10. Braintrust bright data0.98
11. B1.00
12. Browserbase1.00
13. docker1.00
14. :neo4j0.92
15. ORACLE1.00
16. PayPal1.00
17. qodo1.00
18. reducto1.00
19. Sonar1.00
20. Makers of1.00
21. togetherai1.00
22. Unblocked1.00
23. WorkOS1.00
24. SonarQube1.00
0:37 #3 done12 line(s)

shot 3·sharpness 2161.7
1. AlEngineer0.99
2. AI ENGINEER0.98
3. World'sFair1.00
4. World's Fair1.00
5. PRESENTED BY1.00
6. Let's Integrate AI Agents in Event-Souaced0.97
7. Microsoft1.00
8. Systems0.93
9. in0.99
10. ) divakar- kumar0.96
11. Engineering the future of Al0.99
12. World'sFair1.00
0:43 #4 skipped

shot 4·duplicate of #3
1:14 #5 done13 line(s)

shot 5·sharpness 2068.7
1. AlEngineer0.99
2. AI ENGINEER0.98
3. World'sFair1.00
4. World's Fair1.00
5. PRESENTED BY1.00
6. Let's Integrate AI Agents in Event-Souaced0.95
7. Microsoft1.00
8. Systems0.99
9. in0.99
10. @ divakar-kumar0.95
11. TRACK 3· JULY 2, 20260.96
12. Al in Finance0.98
13. World'sFair1.00
1:48 #6 skipped

shot 6·duplicate of #5
2:21 #7 skipped

shot 7·duplicate of #5
2:57 #8 skipped

shot 8·duplicate of #5
3:07 #9 skipped

shot 9·duplicate of #5
3:33 #10 done6 line(s)

shot 10·sharpness 751.0
1. AlEngineer0.99
2. World's Fair0.97
3. Real- i Frannes detection0.65
4. TRACK 3 · JULY 2, 20260.93
5. Al in Finance0.99
6. World'sFair1.00
3:41 #11 done10 line(s)

shot 11·sharpness 1252.4
1. AlEngineer0.99
2. World'sFair1.00
3. Rule Based Appaoach0.97
4. Rales Engine0.90
5. 81.00
6. Fraud0.94
7. Not Frand0.96
8. TRACK 3• JULY 2, 20260.96
9. Al in Finance0.98
10. World's Fair0.96
4:23 #12 done15 line(s)

shot 12·sharpness 1322.5
1. AlEngineer0.97
2. World's Fair0.92
3. Traditina ML Apoach0.92
4. 0.000.82
5. 81.00
6. ML Model0.97
7. Frand Scare0.96
8. Feature1.00
9. Extraction0.99
10. Brand0.90
11. Not Frand0.96
12. TRACK 3· JULY 2, 20260.96
13. Al in Finance0.96
14. World'sFair1.00
15. AlEngineer0.99
4:42 #13 done11 line(s)

shot 13·sharpness 1007.7
1. AlEngineer0.99
2. World'sFair1.00
3. GRay0.75
4. Low0.95
5. HIGt0.82
6. RIsK0.84
7. ZoNE0.76
8. RJSK0.87
9. TRACK 3· JULY 2, 20260.96
10. Al in Finance0.98
11. World'sFair1.00
5:35 #14 done23 line(s)

shot 14·sharpness 3345.2
1. AlEngineer0.98
2. l'ien 1 :0.72
3. Rule Based.0.96
4. Traditional ML Based0.98
5. World'sFair1.00
6. Risk Scone0.89
7. 0-100%0.99
8. PRESENTED BY1.00
9. >80%0.96
10. Microsoft1.00
11. ∠20'%0.88
12. 20-80 %0.89
13. AProve0.88
14. Escalate1.00
15. Reject1.00
16. ~70%0.88
17. ~15.1.0.84
18. ~15%0.94
19. Agantic AI Processing0.99
20. l'ien 2 :0.87
21. TRACK 3· JULY 2, 20260.96
22. Al in Finance0.96
23. World'sFair1.00
5:56 #15 done15 line(s)

shot 15·sharpness 2073.6
1. AlEngineer0.97
2. World'sFair1.00
3. Transaction0.93
4. Context0.99
5. PRESENTED BY1.00
6. Microsoft1.00
7. Payment0.97
8. Context1.00
9. Account's0.99
10. Context1.00
11. Device1.00
12. Context1.00
13. TRACK 3· JULY 2, 20260.96
14. Al in Finance0.96
15. World'sFair1.00
6:22 #16 skipped

shot 16·duplicate of #15
7:03 #17 skipped

shot 17·duplicate of #15
7:37 #18 skipped

shot 18·duplicate of #15
8:13 #19 done17 line(s)

shot 19·sharpness 2272.7
1. AlEngineer0.99
2. World's Fair0.98
3. Transaction0.95
4. Context1.00
5. Message1.00
6. Pay ment0.91
7. Agentic Stone0.99
8. Context1.00
9. Oachestiuton0.86
10. Accounit's0.94
11. Baoken0.85
12. Context1.00
13. Device1.00
14. Context1.00
15. TRACK 3· JULY 2, 20260.96
16. Al in Finance1.00
17. World's Fair0.99
8:23 #20 done20 line(s)

shot 20·sharpness 1920.4
1. AlEngineer0.99
2. World's Fair0.98
3. 1rausaction Context0.95
4. ∠< aggnegate root>>0.92
5. Transaction1.00
6. << domain event >>0.98
7. Transaction Created0.97
8. (<integnation evar>0.89
9. <<integnation evar>0.92
10. (<integnation evour>>0.86
11. Transaction Rajected0.97
12. Transaction Approved0.96
13. Payment Completed0.96
14. (<integration evat>0.91
15. Payment Failed0.94
16. Event Log0.96
17. (transaction-Cvens)0.97
18. TRACK 3· JULY 2, 20260.96
19. Al in Finance0.99
20. World's Fair0.99
8:51 #21 skipped

shot 21·duplicate of #20
9:23 #22 done22 line(s)

shot 22·sharpness 3098.9
1. AlEngineer1.00
2. World'sFair1.00
3. Timeline Projections0.99
4. Customeg activity0.86
5. Risk view1.00
6. Projections0.93
7. Transaction0.92
8. Context1.00
9. CDC0.99
10. ( change F=eed )0.95
11. Event Stream0.99
12. E,0.86
13. E20.97
14. E30.77
15. E40.99
16. Es0.78
17. Use>0.70
18. Command1.00
19. TRACK 3· JULY 2, 20260.96
20. Al in Finance0.98
21. AlEngineer0.95
22. World's Fair0.98
10:00 #23 skipped

shot 23·duplicate of #22

Transcript

204 cues· 3,302 words· 18,068 chars

chunk 0 · 0:13–0:50 · 112 words · 583 chars
0:13 Hello, everyone.
0:15 Thanks for joining.
0:16 So I think finally we are at the last day of the conference.
0:21 Personally, I had a great experience, learned a lot of new things.
0:25 So I believe by the end of the session, you will have at least few key takeaways that you would apply in your work projects.
0:33 So what is it we are going to learn?
0:35 So we are going to learn how to integrate AI agents in your existing system.
0:40 So this system, it could be an event source system, or it could be an event-driven system, or it could be any architecture that your business has invested over the last few years.
chunk 1 · 0:52–1:33 · 111 words · 583 chars
0:52 Because I always believe that these AI agents are not just for the chatbots or the coding assistants.
0:59 So the real value that you could bring out of these AI agents is when you start to apply these into your business workflows.
1:07 And that is what we are going to learn.
1:09 And the problem space that we are going to work on today is the real-time fraud reduction.
1:14 So let me start with an example.
1:15 So this is my personal experience.
1:18 So exactly a month ago, I decided to purchase this laptop that I'm using for the presentation.
chunk 2 · 1:26–2:09 · 126 words · 642 chars
1:26 So it costed me around $3,500.
1:30 So I waited for a right moment.
1:33 So I was seeing whether I have a nice office, and there was one moment.
1:38 So I decided to buy.
1:39 I provided my car details, and then I clicked on Buy button.
1:43 Then my transaction got declined.
1:45 So I thought it was a network issue, and some information has been misplaced.
1:50 So I tried the second attempt, and it was failed again.
1:54 Before I tried the third of them, so I got a call from the customer service asking to verify if I'm doing that particular transaction.
chunk 3 · 2:02–2:46 · 135 words · 735 chars
2:02 And I was like, yes, I was trying to do this for the past few minutes.
2:06 And I asked them, why did you block my transaction?
2:09 Do you know what the response was?
2:12 They didn't know because I wouldn't blame them because they didn't know like why it was blocked.
2:18 It was somewhere in the system, either the rule-based engine or the ML-based engine would have taken that decision.
2:24 So it would have looked through my transaction history or it would have have an average threshold beyond which like if it goes like just block the transaction, that would be the static rule that it would have had because of which my transaction got declined.
chunk 4 · 2:39–3:19 · 118 words · 638 chars
2:39 So those are the key areas, those are the uncertain areas where we are trying to integrate the AI agents.
2:46 Now, you might be thinking, what a great idiot, right?
2:50 Because it is already an uncertain case.
2:53 Why do you want to introduce an AI agent?
2:55 Because it is also a non-deterministic by nature, right?
2:58 But the key point here that we are all trying to miss is that earlier in the rule-based engine or the ML-based engine, we don't have enough context.
3:08 We don't have enough real-time data that gets passed on to the system.
chunk 5 · 3:13–3:55 · 125 words · 675 chars
3:13 And those are the real data that we are trying to capture from different bounded contexts that we have in our domain.
3:19 And we are going to see how we could build that architecture so that the AI agent can make use of it and come up with a verdict.
3:30 So the domain that we are gonna talk about is the real-time fraud detection.
3:34 As I mentioned before, so we had this rule-based engine like five years before, and this rule-based engine was perfectly fine, like it was working perfectly fine for a few of the cases.
chunk 6 · 3:45–4:29 · 122 words · 658 chars
3:45 But the problem with this rule-based engine is like the maintainability, because the fraudsters are,
3:51 trying to get intruded into a system by a lot of different ways.
3:56 And you just need to keep on updating these static rules day by day.
3:59 And it's going to be really difficult for you to manage.
4:03 And that's when we started to tie up with a third party provider who helped us to develop this ML model.
4:10 So we had this ML-based approach where
4:15 shared with them transaction history or different features with them based on that.
chunk 7 · 4:21–5:05 · 122 words · 689 chars
4:21 They trained the ML model, and we were able to get a risk code based on that, with which we were able to block or approve the transaction.
4:30 But the problem with either of these approaches, like either, like we were able to handle most of the transaction because it would fall below a certain threshold, then we would approve the transaction.

Chapters

0:00 Introduction: adding agents to an existing system
1:20 A declined transaction you can't explain
3:04 Where rule based and ML systems fall short
5:40 Handling the gray zone with agents
5:53 Bounded contexts: transaction, device, account
8:24 Event sourcing and change feeds
10:57 Building the semantic layer
13:16 Avoiding infinite loops
14:07 The risk analyzer and verdict agents
15:24 The saga orchestration loop
19:00 Putting the architecture together