vidtheque.

read-only demo

Videos z0sh8HyTrDo

Your Finance Agent's Bottleneck Is You — Ramana Siddanth Emani, Auditoria AI

index_state ready data_status ok

AI Engineer· published 2026-07-30· 0:13:42· en-US· indexed 2026-08-10 19:38

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:03, 1 of 1 keyframes kept
  2. Shot 1, 0:03 to 0:05, 1 of 1 keyframes kept
  3. Shot 2, 0:05 to 0:12, 1 of 1 keyframes kept
  4. Shot 3, 0:12 to 0:16, 1 of 1 keyframes kept
  5. Shot 4, 0:16 to 0:22, 1 of 1 keyframes kept
  6. Shot 5, 0:22 to 1:09, 1 of 1 keyframes kept
  7. Shot 6, 1:09 to 1:48, 1 of 1 keyframes kept
  8. Shot 7, 1:48 to 2:21, 1 of 1 keyframes kept
  9. Shot 8, 2:21 to 2:54, 0 of 1 keyframes kept
  10. Shot 9, 2:54 to 3:05, 1 of 1 keyframes kept
  11. Shot 10, 3:05 to 3:33, 1 of 1 keyframes kept
  12. Shot 11, 3:33 to 4:00, 0 of 1 keyframes kept
  13. Shot 12, 4:00 to 4:28, 0 of 1 keyframes kept
  14. Shot 13, 4:28 to 5:13, 1 of 1 keyframes kept
  15. Shot 14, 5:13 to 5:45, 1 of 1 keyframes kept
  16. Shot 15, 5:45 to 6:18, 0 of 1 keyframes kept
  17. Shot 16, 6:18 to 6:51, 1 of 1 keyframes kept
  18. Shot 17, 6:51 to 7:23, 1 of 1 keyframes kept
  19. Shot 18, 7:23 to 7:54, 1 of 1 keyframes kept
  20. Shot 19, 7:54 to 8:26, 1 of 1 keyframes kept
  21. Shot 20, 8:26 to 8:58, 0 of 1 keyframes kept
  22. Shot 21, 8:58 to 9:35, 1 of 1 keyframes kept
  23. Shot 22, 9:35 to 10:07, 1 of 1 keyframes kept
  24. Shot 23, 10:07 to 10:39, 1 of 1 keyframes kept
  25. Shot 24, 10:39 to 11:04, 1 of 1 keyframes kept
  26. Shot 25, 11:04 to 11:29, 0 of 1 keyframes kept
  27. Shot 26, 11:29 to 11:55, 1 of 1 keyframes kept
  28. Shot 27, 11:55 to 12:20, 0 of 1 keyframes kept
  29. Shot 28, 12:20 to 12:46, 0 of 1 keyframes kept
  30. Shot 29, 12:46 to 13:13, 0 of 1 keyframes kept
  31. Shot 30, 13:13 to 13:18, 1 of 1 keyframes kept
  32. Shot 31, 13:18 to 13:25, 1 of 1 keyframes kept
  33. Shot 32, 13:25 to 13:41, 0 of 1 keyframes kept

33 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
127
whisperx 127
chunks
24
from 127 cues
keyframes
23
kept of 33 captured
frames with text
23
324 lines read
chapters
11
from the source metadata
keyframe bytes
3.2 MB
word timings on 127 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-09 21:54 0s
stt done — 2026-08-09 03:07 15s
chunk done — 2026-08-09 03:08 0s
text_embed done — 2026-08-10 19:38 0s
keyframe done — 2026-08-09 03:08 1m 42s
ocr done — 2026-08-09 03:09 7s
frame_embed done — 2026-08-10 19:38 4s

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

  • 0:13 #3 done2 line(s)

    shot 3·sharpness 302.2

    1. AlEngineer1.00
    2. World's Fair0.98

  • 0:17 #4 done2 line(s)

    shot 4·sharpness 43.2

    1. AlEngineer0.99
    2. World's Fair0.99

  • 0:36 #5 done13 line(s)

    shot 5·sharpness 2087.6

    1. AlEngineer0.96
    2. World'sFair1.00
    3. AI ENGINEER WORLD'S FAIR 20260.98
    4. HARNESS ENGINEERING1.00
    5. AI FOR FINANCE1.00
    6. YOUR FINANCE AGENT'S0.99
    7. PRESENTED BY1.00
    8. BOTTLENECK IS YOU0.97
    9. Microsoft1.00
    10. The developer harness that ships production multi-agent systems - not demos.1.00
    11. ./ship --past-pilot # Siddanth Emani · Auditoria AI0.95
    12. Engineering the future of Al1.00
    13. World'sFair0.97

  • 1:14 #6 done19 line(s)

    shot 6·sharpness 1690.8

    1. AlEngineer0.98
    2. World'sFair1.00
    3. Demos ship. Pilots don't.1.00
    4. DEMO1.00
    5. looks like magic1.00
    6. PRESENTED BY1.00
    7. Most "AI for finance" demos look great.1.00
    8. Microsoft1.00
    9. Almost none survive past pilot.0.99
    10. PILOT1.00
    11. meets real AP data1.00
    12. The demo gods are generous.0.98
    13. PROD1.00
    14. x most die here0.97
    15. The production gods bill by the hour.1.00
    16. GRAVEYARD1.00
    17. TRACK 3· JULY 2,20260.94
    18. Al in Finance0.99
    19. World'sFair0.96

  • 2:17 #7 done23 line(s)

    shot 7·sharpness 1145.5

    1. AlEngineer0.98
    2. World's Fair0.96
    3. > It's not what you'd guess0.95
    4. Model capability vs. your dev-loop velocity1.00
    5. MODEL1.00
    6. commodity - swap it, nothing changes0.99
    7. 1201.00
    8. 1001.00
    9. 801.00
    10. COMPUTE1.00
    11. 601.00
    12. faster GPUs don't fix a stuck pilot0.99
    13. 401.00
    14. 201.00
    15. 01.00
    16. 21.00
    17. FRAMEWORK1.00
    18. agents, MCP - that's plumbing0.97
    19. Model capability0.97
    20. Your dev loop1.00
    21. TRACK 3· JULY 2, 20260.93
    22. Alin Finance0.99
    23. World's Fair0.99

  • 2:41 #8 skipped

    shot 8·duplicate of #7

  • 2:55 #9 done9 line(s)

    shot 9·sharpness 947.5

    1. AlEngineer0.98
    2. World'sFair1.00
    3. 11.00
    4. ACT I0.91
    5. The Harness1.00
    6. Four Primitives. Think Loops. 10x Production Code.1.00
    7. TRACK 3· JULY 2, 20260.96
    8. Alin Finance0.98
    9. World'sFair1.00

  • 3:27 #10 done21 line(s)

    shot 10·sharpness 1454.7

    1. AlEngineer0.98
    2. World's Fair0.99
    3. >0.70
    4. Loop1.00
    5. Primitives1.00
    6. 由0.95
    7. 011.00
    8. Subagent fleet1.00
    9. Parallel Worktrees → FAN OUT, don't queue1.00
    10. 021.00
    11. Skills1.00
    12. Secret Recipes → Run Forever0.99
    13. 031.00
    14. MCP pipeline1.00
    15. Prod Failures → Traces → Fixes0.97
    16. 041.00
    17. Minimal UX1.00
    18. All the above · Tickets · PRs · logs → one window0.95
    19. TRACK 3· JULY 2, 20260.93
    20. Al in Finance0.99
    21. World's Fair0.99

  • 3:36 #11 skipped

    shot 11·duplicate of #10

  • 4:11 #12 skipped

    shot 12·duplicate of #10

  • 4:41 #13 done32 line(s)

    shot 13·sharpness 1330.6

    1. AlEngineer0.98
    2. World'sFair1.00
    3. SUBAGENTS1.00
    4. Fan out, don't queue up1.00
    5. wt-011.00
    6. wt-021.00
    7. wt-031.00
    8. v done0.99
    9. running1.00
    10. v done0.99
    11. Every task gets its own git worktree.0.99
    12. A main agent works each tree in parallel while you steer.1.00
    13. wt-040.99
    14. wt-050.93
    15. wt-860.91
    16. running1.00
    17. needs you1.00
    18. v done1.00
    19. you1.00
    20. wt-071.00
    21. wt-080.94
    22. wt-090.99
    23. 501.00
    24. running1.00
    25. v done0.98
    26. running1.00
    27. active worktrees1.00
    28. 1 orchestrator1.00
    29. "My laptop fans now qualify as a wind tunnel0.99
    30. TRACK 3· JULY 2, 20260.93
    31. Al in Finance0.98
    32. World'sFair1.00

  • 5:35 #14 done20 line(s)

    shot 14·sharpness 1747.5

    1. AlEngineer0.98
    2. World's Fair0.98
    3. >0.66
    4. /jira-to-qa - the bug-fix harness0.99
    5. Parse requirements1.00
    6. Pull trace (MCP)1.00
    7. PRESENTED BY0.96
    8. RCA + checkpoint0.95
    9. Microsoft1.00
    10. Worktree1.00
    11. TDD the fix1.00
    12. Local e2e1.00
    13. PR + merge1.00
    14. 81.00
    15. Build + dev1.00
    16. 91.00
    17. Stage + validate1.00
    18. TRACK 3· JULY 2, 20260.95
    19. Al in Finance0.99
    20. World's Fair0.98

  • 5:52 #15 skipped

    shot 15·duplicate of #14

  • 6:28 #16 done18 line(s)

    shot 16·sharpness 1616.8

    1. AlEngineer0.99
    2. World's Fair0.97
    3. 70.52
    4. /jira-to-qa - the bug-fix harness0.99
    5. Parse requirements1.00
    6. Pull trace (MCP)1.00
    7. RCA + checkpoint0.97
    8. Worktree1.00
    9. TDD the fix1.00
    10. Local e2e1.00
    11. PR + merge1.00
    12. 81.00
    13. Build + dev1.00
    14. 91.00
    15. Stage + validate0.99
    16. TRACK 3· JULY 2, 20260.96
    17. Al in Finance0.99
    18. World's Fair0.97

  • 7:04 #17 done15 line(s)

    shot 17·sharpness 1432.5

    1. AlEngineer0.98
    2. World'sFair1.00
    3. MINIMAL UX1.00
    4. >0.54
    5. One pane of glass0.99
    6. Collapses tickets, PRs, logs & deploys0.99
    7. into one always-on window.1.00
    8. Neck rotations to ship one change1.00
    9. 471.00
    10. Chrome Tabs & Application Windows1.00
    11. before the widget = 47.0.99
    12. after the widget = 10.97
    13. TRACK 3· JULY 2, 20260.93
    14. Al in Finance0.95
    15. World'sFair1.00

  • 7:39 #18 done14 line(s)

    shot 18·sharpness 1431.2

    1. AlEngineer0.98
    2. World'sFair1.00
    3. MINIMAL UX1.00
    4. One pane of glass0.97
    5. Collapses tickets, PRs, logs & deploys1.00
    6. into one always-on window.1.00
    7. Neck rotations to ship one change1.00
    8. 471.00
    9. Chrome Tabs & Application Windows0.99
    10. before the widget = 47.1.00
    11. after the widget = 10.97
    12. TRACK 3· JULY 2, 20260.93
    13. Al in Finance0.99
    14. World'sFair1.00

  • 8:13 #19 done21 line(s)

    shot 19·sharpness 1786.7

    1. AlEngineer0.98
    2. World's Fair0.99
    3. Verification is regulatory, not a test suite1.00
    4. Software1.00
    5. Finance1.00
    6. Reviewer1.00
    7. CI / another agent0.96
    8. a human auditor0.98
    9. “Pass" means0.96
    10. tests are green1.00
    11. controller signs off under SOX1.00
    12. Review agent → agent?0.98
    13. yes1.00
    14. accountability?1.00
    15. The bottleneck1.00
    16. the merge queue1.00
    17. human attention1.00
    18. "Move fast and break things" lands differently when an auditor reads it.0.98
    19. TRACK 3· JULY 2, 20260.94
    20. Al in Finance1.00
    21. World's Fair0.98

  • 8:51 #20 skipped

    shot 20·duplicate of #19

  • 9:06 #21 done9 line(s)

    shot 21·sharpness 1139.2

    1. AlEngineer0.98
    2. World'sFair1.00
    3. 21.00
    4. PART II1.00
    5. Removing yourself from the loop1.00
    6. the frontier - and recursive loops0.98
    7. TRACK 3· JULY 2, 20260.95
    8. Al in Finance0.98
    9. World'sFair1.00

  • 9:57 #22 done15 line(s)

    shot 22·sharpness 1589.9

    1. AlEngineer0.98
    2. World'sFair1.00
    3. Production failures become input1.00
    4. Production Bugs1.00
    5. Traces (MCP)0.98
    6. Tickets1.00
    7. /jira-to-qa0.99
    8. a user flags a bad answer1.00
    9. the exact failing run1.00
    10. auto-drafted from evidence1.00
    11. fix → deploy → verify0.93
    12. Production Failures stop being fire drills and become typed input into your agents.1.00
    13. TRACK 3· JULY 2,20260.96
    14. Alin Finance0.96
    15. World'sFair1.00

  • 10:35 #23 done18 line(s)

    shot 23·sharpness 1729.0

    1. AlEngineer0.99
    2. World'sFair1.00
    3. 70.51
    4. Production failures become input1.00
    5. PRESENTED BY1.00
    6. Microsoft1.00
    7. Production Bugs1.00
    8. Traces (MCP)0.99
    9. Tickets1.00
    10. /jira-to-qa0.99
    11. a user flags a bad answer1.00
    12. the exact failing run1.00
    13. auto-drafted from evidence0.98
    14. fix → deploy → verify0.94
    15. Production Failures stop being fire drills and become typed input into your agents.1.00
    16. TRACK 3· JULY 2, 20260.96
    17. Al in Finance0.95
    18. World'sFair1.00

Transcript

127 cues· 1,887 words· 10,106 chars

  1. 0:12 Hello, everyone.
  2. 0:13 Welcome to this session about your finance agent's bottleneck is you.
  3. 0:19 So sorry for the rude title.
  4. 0:21 I don't mean to call the audience here the bottlenecks, but I'm here to talk about the harnesses that you guys are developing and using these internal harnesses to build your production agents.
  5. 0:34 So my name is Siddhant Timani, and I'm a data scientist at Auditorio AI, and we build production agents for finance.
  6. 0:44 So if you're a CFO in the audience, I would love to speak to you after the session.
  7. 0:50 This talk is in between the harness engineering track and AI for finance.
  8. 0:55 So this talk is mostly about identifying the bottlenecks within your developer harnesses,
  9. 1:01 And if you're a developer yourself, how do you be 10x productive with the agent harnesses that you're using?
  10. 1:10 So all of us have seen beautiful demos in this AI engineers' world fair.
  11. 1:16 But once these demos are promoted to pilots and you start onboarding new customers, the agent has never seen these future data.
  12. 1:29 All of us know production bugs are very high, and production gods build by the hour.
  13. 1:36 So that's a hard fact.
  14. 1:39 And writing code is very easy.
  15. 1:41 So shipping beautiful demos and showing it to a lot of people is very easy nowadays.
  16. 1:49 So what is the problem?
  17. 1:51 And why do these demos fail in production?
  18. 1:54 Is it the model?
  19. 1:56 Do you need a better model, Fable 5, perhaps?
  20. 2:00 Or do you need faster GPUs?
  21. 2:02 Or do you need a better framework, maybe?
  22. 2:06 Or your Ralph loops are not working properly?
  23. 2:10 So what is the answer?
  24. 2:13 If we wait three and a half months, we are rewarded with a new model in the market.
  25. 2:17 So we can easily swap models.
  26. 2:20 If we wait perhaps one year, we have new chips.
  27. 2:23 We have faster GPUs.
  28. 2:26 And again, writing code is easy.
  29. 2:28 So we have new frameworks every day.
  30. 2:31 So you can swap your framework every now and then.
  31. 2:35 So how do we, in real time, fix these production bugs?
  32. 2:40 The answer is your developed velocity.
  33. 2:43 The model capability increases very exponentially.
  34. 2:46 And the developers have to spend a lot of time
  35. 2:50 every day to automate your developer loop.
  36. 2:55 So I'm talking about four primitives here.
  37. 2:59 All of you need to think about loops.
  38. 3:01 And at the end of the session, I hope you can 10X your production code.
  39. 3:07 So first we have sub-agents.
  40. 3:10 Nowadays, whatever harness you're using, you can spawn new sub-agents.
  41. 3:15 You can have an army of them.
  42. 3:17 and Git Worktrees are your best friend.
  43. 3:21 So think of Worktrees as isolated folders, and inside these folders, the agent writes whatever code it's generating.
  44. 3:29 So you want these Worktrees to be in parallel so that subagents are doing independent tasks and are not fighting over the same thing.
  45. 3:39 Second, we have skills.
  46. 3:40 These are your organization's secret recipes.
  47. 3:43 So make sure you have a lot of skills because these skills, once you start giving it to your agents, the agents will always make sure to use the correct and proper workflows to solve whatever production bug you're facing.
  48. 4:00 And of course, all of us have seen a lot of MCP tools being shipped into the market right now.
  49. 4:05 Everybody says the agent can connect to whatever MCP tool and whatever third-party server there is.
  50. 4:12 And your client data can live in any system you want.

Chapters

  1. 0:00 Your bottleneck is you
  2. 1:05 From bugs to pilots to production
  3. 2:37 Automating the developer loop
  4. 3:03 Coding agents that multiply output
  5. 3:39 Skills for the right patterns
  6. 4:22 Sub agents and where tasks come from
  7. 5:18 Pulling traces, testing, reporting back
  8. 7:56 Auditoria in the finance sector
  9. 9:04 Stepping out of the loop safely
  10. 11:35 Turning customer patterns into features
  11. 13:04 Keep a human as verifier

Open at this second