vidtheque.

read-only demo

Videos RGiXcVxSD3s

Chat and citations won't save your vertical AI - Atul Ramachandran, Filed Inc

index_state ready data_status ok

AI Engineer· published 2026-07-11· 0:15:12· en-US· indexed 2026-08-11 04:05

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:03, 1 of 1 keyframes kept
  2. Shot 1, 0:03 to 0:52, 1 of 1 keyframes kept
  3. Shot 2, 0:52 to 1:01, 1 of 1 keyframes kept
  4. Shot 3, 1:01 to 1:37, 1 of 1 keyframes kept
  5. Shot 4, 1:37 to 2:06, 1 of 1 keyframes kept
  6. Shot 5, 2:06 to 2:35, 0 of 1 keyframes kept
  7. Shot 6, 2:35 to 3:04, 0 of 1 keyframes kept
  8. Shot 7, 3:04 to 3:34, 1 of 1 keyframes kept
  9. Shot 8, 3:34 to 4:04, 0 of 1 keyframes kept
  10. Shot 9, 4:04 to 4:34, 0 of 1 keyframes kept
  11. Shot 10, 4:34 to 5:04, 0 of 1 keyframes kept
  12. Shot 11, 5:04 to 5:31, 1 of 1 keyframes kept
  13. Shot 12, 5:31 to 5:58, 0 of 1 keyframes kept
  14. Shot 13, 5:58 to 6:21, 1 of 1 keyframes kept
  15. Shot 14, 6:21 to 7:03, 1 of 1 keyframes kept
  16. Shot 15, 7:03 to 7:30, 1 of 1 keyframes kept
  17. Shot 16, 7:30 to 7:57, 0 of 1 keyframes kept
  18. Shot 17, 7:57 to 8:24, 0 of 1 keyframes kept
  19. Shot 18, 8:24 to 8:51, 1 of 1 keyframes kept
  20. Shot 19, 8:51 to 9:18, 0 of 1 keyframes kept
  21. Shot 20, 9:18 to 9:45, 0 of 1 keyframes kept
  22. Shot 21, 9:45 to 10:12, 1 of 1 keyframes kept
  23. Shot 22, 10:12 to 10:39, 0 of 1 keyframes kept
  24. Shot 23, 10:39 to 11:12, 1 of 1 keyframes kept
  25. Shot 24, 11:12 to 11:45, 0 of 1 keyframes kept
  26. Shot 25, 11:45 to 12:13, 1 of 1 keyframes kept
  27. Shot 26, 12:13 to 12:41, 0 of 1 keyframes kept
  28. Shot 27, 12:41 to 13:10, 0 of 1 keyframes kept
  29. Shot 28, 13:10 to 13:39, 0 of 1 keyframes kept
  30. Shot 29, 13:39 to 14:08, 1 of 1 keyframes kept
  31. Shot 30, 14:08 to 14:37, 0 of 1 keyframes kept
  32. Shot 31, 14:37 to 15:05, 1 of 1 keyframes kept
  33. Shot 32, 15:05 to 15:11, 1 of 1 keyframes kept

33 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
167
whisperx 167
chunks
28
from 167 cues
keyframes
17
kept of 33 captured
frames with text
17
312 lines read
chapters
0
from the source metadata
keyframe bytes
3.1 MB
word timings on 167 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-11 04:02 1m 06s
stt done — 2026-08-11 04:03 16s
chunk done — 2026-08-11 04:03 0s
text_embed done — 2026-08-11 04:03 1s
keyframe done — 2026-08-11 04:03 1m 15s
ocr done — 2026-08-11 04:04 10s
frame_embed done — 2026-08-11 04:05 4s

Frames, and what the machine read

  • 0:02 #0 done5 line(s)

    shot 0·sharpness 962.8

    1. Filed.1.00
    2. AI ENGINEER 20260.98
    3. Chat and citations won't save your1.00
    4. verticalAI.1.00
    5. Atul Ramachandran, CTO & Co-founder, Filed0.99

  • 0:33 #1 done3 line(s)

    shot 1·sharpness 904.0

    1. THE THESIS1.00
    2. Chat and citations are the current answer.1.00
    3. Here's why they break that promise.0.98

  • 1:00 #2 done3 line(s)

    shot 2·sharpness 649.8

    1. WHOIAM1.00
    2. Atul Ramachandran1.00
    3. CTO & Co-founder, Filed1.00

  • 1:33 #3 done7 line(s)

    shot 3·sharpness 716.8

    1. STARTUPS1.00
    2. Filed.1.00
    3. AI for tax1.00
    4. Filed raises $17M to automate1.00
    5. the drudgery of tax prep1.00
    6. professionals.1.00
    7. FOUNDED 20241.00

  • 2:00 #4 done32 line(s)

    shot 4·sharpness 2962.3

    1. What were the most common tax law changes affecting 1040 returns this year?0.99
    2. Give me the IRS links only. Make no mistakes0.98
    3. WHAT CHAT AND CITATIONS1.00
    4. ARE1.00
    5. If you're talking about 2025 tax year returns filed during the 2026 filing season, these were the most common 1040-impacting0.99
    6. changes according to IRS guidance:1.00
    7. • Standard deduction changes and inflation adjustments0.97
    8. IRS Form 1040 Instructions (2025)0.99
    9. • Child Tax Credit changes0.96
    10. IRS Form 1040 Instructions (2025)1.00
    11. • New Schedule 1-A for additional deductions0.98
    12. IRS Schedule 1-A Overview0.98
    13. Chat.1.00
    14. • Additional deduction for taxpayers age 65+0.99
    15. IRS Seniors Filing Season Updates0.98
    16. • Deduction for qualified tips0.98
    17. Citations.1.00
    18. IRS One Big Beautiful Bill Provisions (Individuals and Workers)0.99
    19. • Deduction for qualified overtime pay0.99
    20. IRS One Big Beautiful Bill Provisions (Individuals and Workers)0.99
    21. The default UX for AI agents.0.99
    22. • Deduction for qualified car loan interest0.98
    23. IRS One Big Beautiful Bill Provisions (Individuals and Workers)0.99
    24. • Increased SALT deduction limit0.97
    25. IRS Form 1040 Instructions (2025)1.00
    26. • IRS summary of all major individual tax law changes from the 2025 legislation0.99
    27. IRS One Big Beautiful Bill Provisions0.99
    28. Source verification: all links above are official IRS pages and refler hanges affecting 2025 returns.0.99
    29. ②0.66
    30. ..Sources0.95
    31. Ask anything1.00
    32. + ⊕ A Auto0.81

  • 2:18 #5 skipped

    shot 5·duplicate of #4

  • 3:01 #6 skipped

    shot 6·duplicate of #4

  • 3:31 #7 done8 line(s)

    shot 7·sharpness 1203.8

    1. A HISTORY OF PRODUCT ABSTRACTIONS1.00
    2. Three layers of abstraction. Throughout history.0.98
    3. O3 Agentic delegation0.98
    4. You don't use the product. You delegate to it. You are not doing the work.0.99
    5. O2 Digital transformation0.97
    6. The bank became online banking. The grocery store became an app.0.99
    7. olPhysical0.97
    8. A bank branch. You showed up in person.0.98

  • 3:44 #8 skipped

    shot 8·duplicate of #7

  • 4:31 #9 skipped

    shot 9·duplicate of #7

  • 5:01 #10 skipped

    shot 10·duplicate of #7

  • 5:13 #11 done12 line(s)

    shot 11·sharpness 1610.0

    1. THE CONVEYOR BELT0.99
    2. The product is the conveyor belt. The users are the supervisors.0.99
    3. You, the supervisor delegate·instruct·decide0.99
    4. ↓0.56
    5. sends a task0.97
    6. new task0.98
    7. done1.00
    8. Agent1.00
    9. Agent1.00
    10. Agent1.00
    11. Agent1.00
    12. The agents do the work.1.00

  • 5:45 #12 skipped

    shot 12·duplicate of #11

  • 6:05 #13 done4 line(s)

    shot 13·sharpness 911.5

    1. HOW DO WE BUILD THE CONVEYOR BELT1.00
    2. If a user wants to delegate instead of do, what does that interface look0.99
    3. like?1.00
    4. Design for delegation, not participation.1.00

  • 6:42 #14 done16 line(s)

    shot 14·sharpness 999.5

    1. FOUR COMPONENTS0.99
    2. Every agentic product needs four things: delegate, teach, monitor, intervene.1.00
    3. 010.77
    4. O20.83
    5. 040.92
    6. Delegate1.00
    7. Teach1.00
    8. Monitor1.00
    9. Intervene1.00
    10. Hand off a task to the agent.0.99
    11. Teach the agent how a task1.00
    12. See where any task is at any0.99
    13. Step in for errors or judgment0.99
    14. should be done.0.97
    15. time.1.00
    16. calls.1.00

  • 7:06 #15 done12 line(s)

    shot 15·sharpness 824.5

    1. COMPONENT O1·DELEGATE0.99
    2. Delegation.1.00
    3. Identify the tasks that take a couple of hours, and build background agents.0.99
    4. Taxprep1.00
    5. Tax review0.96
    6. Taxplan1.00
    7. takes1.00
    8. takes1.00
    9. takes1.00
    10. >2hrs1.00
    11. >2hrs1.00
    12. >4hrs1.00

  • 7:51 #16 skipped

    shot 16·duplicate of #1

  • 8:00 #17 skipped

    shot 17·duplicate of #15

  • 8:33 #18 done7 line(s)

    shot 18·sharpness 663.4

    1. COMPONENT O2· TEACH0.96
    2. Teach (Skills)0.97
    3. Predefined agents get you 80% of the way. Skills close the last 20%: the quirks.1.00
    4. 80%0.97
    5. 20%1.00
    6. Predefined background agents1.00
    7. Taught with skills1.00

  • 9:00 #19 skipped

    shot 19·duplicate of #18

  • 9:34 #20 skipped

    shot 20·duplicate of #18

  • 10:04 #21 done45 line(s)

    shot 21·sharpness 1447.8

    1. COMPONENT O3·MONITOR0.99
    2. Monitor. Not where the work happens, but where trust is maintained.0.99
    3. Tasis0.83
    4. 102,0000.94
    5. Tasks1.00
    6. OUTPUT-LINE IA WAGES0.91
    7. Track and manage v0.97
    8. okerage Servioes LLC (2025)0.95
    9. 82,40670.77
    10. Atut R0.76
    11. 2025 SUPPLEMENTAL INFORMATION0.98
    12. Task ts0.79
    13. Clent 120.79
    14. Triggered by0.97
    15. Status 140.87
    16. Created at +0.88
    17. Tax Advisor0.89
    18. John Doe0.88
    19. just now0.91
    20. 0Tax Review0.90
    21. John Doe0.94
    22. jast now0.88
    23. =0.53
    24. Binder0.94
    25. Oliver Fied0.94
    26. 6 min ago1.00
    27. D Taxgrep0.89
    28. 5 days ago0.95
    29. O Taxgrep0.85
    30. Oliver Filed0.90
    31. 6 days ago0.94
    32. O Research0.91
    33. Oliver Fied0.93
    34. 7 days ago0.89
    35. okerage Services LLC (2025)0.95
    36. 150.360.99
    37. Chat0.90
    38. Oliver Filed0.90
    39. 7 days ago0.93
    40. Draft emal0.86
    41. Hello0.84
    42. 7 dayt ago0.76
    43. 2025 SUPPLEMENTAL INFORMATION0.98
    44. 7 dmyt sgo0.79
    45. 6 dayt ago0.83

  • 10:31 #22 skipped

    shot 22·duplicate of #21

  • 11:02 #23 done92 line(s)

    shot 23·sharpness 1799.0

    1. Control.1.00
    2. Clients Alex & Jane Filed Binder0.96
    3. Stepping in should feel like1.00
    4. Docs0.99
    5. Forms1.00
    6. Leads1.00
    7. Q 00.53
    8. a0.77
    9. 80.64
    10. …0.54
    11. Notes & flags1.00
    12. Chat1.00
    13. Shortcuts1.00
    14. taking the wheel, not1.00
    15. Filters1.00
    16. 10.57
    17. <0.71
    18. 1099-CONS: Fidelity Brok.0.98
    19. 180.78
    20. V0.69
    21. 100%1.00
    22. +1.00
    23. Q1.00
    24. ←1.00
    25. Thread1.00
    26. abandoning the car.1.00
    27. *1.00
    28. Missing documents1.00
    29. 20251.00
    30. TAX REPORTING1.00
    31. STATEMENT1.00
    32. Agent1.00
    33. Jun 22 - 19:330.97
    34. Observations1.00
    35. 140.94
    36. 1099-B Box B short-term transactions ($4,746.600.99
    37. 三0.69
    38. Source documents1.00
    39. FORM 1099-8°0.95
    40. 2025 Proceeds from Broker and Barter Exchange Transactions1.00
    41. proceeds, cost basis $0.00 Fidelity Brokerage0.99
    42. Services LLC ) involve RSU sales from FALSE0.99
    43. Pause the belt1.00
    44. Income1.00
    45. LANDSCAPE SUPPLY INC. The supplemental stock1.00
    46. 80.87
    47. 1099-CONS: Fidelity Brokerage Servi..0.97
    48. plan lot detail Fidelity Brokerage Services LLC0.97
    49. shows ordinary income was reported to the0.99
    50. V0.81
    51. K-1 (1065): Blue Horizon Ventures LLC1.00
    52. employee ($4,320.63 across three lots) and is1.00
    53. K-1 (1065): Ironclad Manufacturing LL.0.97
    54. already included in W-2 Box 1 wages False0.99
    55. Fix the problem0.98
    56. K-1 (1065): Orion Real Estate Fund LP0.98
    57. FOTALS0.82
    58. 0.00.87
    59. 6.000.92
    60. 0.000.83
    61. ($4,470.63) reflected in the supplemental detail1.00
    62. landscape supply inc The adjusted cost basis0.98
    63. K-1 (1065): Silverline Tech Partners LP0.99
    64. should be used on Form 8949 rather than the $01.00
    65. V0.65
    66. K-1 (1065): Summit Ridge Holdings LP1.00
    67. Rental property income expenses: Lu..D0.96
    68. D0.99
    69. reported to the IRS, to avoid double taxation. CPA0.99
    70. should confirm that W-2 Box 14 RSU amount0.99
    71. ($4,059.72) aligns with supplemental ordinary0.99
    72. Start it back up0.99
    73. Rental property income expenses: M..0.97
    74. income totals before finalizing basis adjustment.1.00
    75. Rental property income expenses: Oc..0.99
    76. Self employment income expenses: S..0.96
    77. @Agent can you please then exclude this0.99
    78. Tax organizer: Alex Filed & Jane Filed0.97
    79. document?1.00
    80. W-2: False landscape supply inc1.00
    81. D0.64
    82. Peon o0.52
    83. Send0.93
    84. Adjustments / Deductions0.97
    85. 1098: Soundview Federal Bank0.99
    86. 20251.00
    87. TAX REPORTING STATEMENT1.00
    88. 8829: Home Office Expense Summary0.99
    89. Excluded / Not Taxable0.99
    90. FORM 1099-8°0.95
    91. Prior year0.99
    92. Upload more documents1.00

Transcript

167 cues· 2,756 words· 14,845 chars

  1. 0:01 Tag and citations won't save you vertical AI.
  2. 0:03 I'm sure many of us are building products in a vertical industry, be it healthcare, legal, or taxes.
  3. 0:12 And usually doing so, we sell one promise to the customers.
  4. 0:15 Either we're going to save you money or save you cost.
  5. 0:18 And usually the pitch goes, AI agents are here.
  6. 0:22 They will do the work for you while you sleep.
  7. 0:25 Now, primary interface that we use for this are chat and citations to interact with an AI agent, right?
  8. 0:31 Chats are usually used for inputs.
  9. 0:33 They allow you to be flexible.
  10. 0:35 You can talk to the agents as you please.
  11. 0:37 While citations are usually used for outputs, they allow you to see the results, verify the results, and so on.
  12. 0:44 Now, I'm here to tell you that citations and chat alone will not allow you to keep the promise that you made to the customers, promise of saving time and money for them.
  13. 0:54 So who am I?
  14. 0:54 I'm Atul.
  15. 0:55 I'm the CTO and co-founder at Fylde.
  16. 0:57 I've been fortunate enough to build products for more than a decade now.
  17. 1:01 At Fylde, we build products for tax professionals in the US.
  18. 1:07 We have raised more than $17 million to do so.
  19. 1:09 And about two years now doing this, we have seen massive growth in the space.
  20. 1:15 Just to give you an idea, just last month, we closed more revenue than what we have done in the one year alone before that.
  21. 1:21 So the growth has been tremendous and we are very excited for this.
  22. 1:25 Doing this entire journey for the last two years, we have learned quite a bit.
  23. 1:30 And I think most of the learnings of building these AI agents for the taxes industry is essentially transferable to any other agent tech product.
  24. 1:37 Now, chat is great.
  25. 1:39 Like it allows us to build fast.
  26. 1:41 It allows customers to interact with products in ways that was previously possible.
  27. 1:45 It's a communication platform, right?
  28. 1:47 You can interact with agents to do tasks, flexible tasks that was not constrained to the UIs that you have built.
  29. 1:53 Citations are also amazing.
  30. 1:55 They ground the answers in truth.
  31. 1:56 Other than allowing the users to verify the outputs, they also serve another purpose.
  32. 2:01 They allow the agents to essentially be more accurate because now they are forced to present the references to their answers.
  33. 2:10 Agents perform much better with citations.
  34. 2:12 So they reduce hallucinations essentially.
  35. 2:16 But there's a key problem here, right?
  36. 2:19 Chat is synchronous.
  37. 2:20 If you think about it, even when you're coding, let's say you're using cloud code, as soon as you're typing the request, once the request is typed, you're waiting for the agent to respond.
  38. 2:31 This synchronous medium does not allow the customers to leave the platform and go and do their work.
  39. 2:37 Similarly, citations also puts the verification burden back into the customer.
  40. 2:42 So think about going and reviewing the agent's work now one by one to ensure everything is correct.
  41. 2:49 Especially in vertical industry space like healthcare, legal, and taxes, this becomes very crucial because now this adds an extra work and customers usually complain, you know.
  42. 2:59 The promise of agents doing the work for me while asleep is not really kept here.
  43. 3:05 Now, before we dive in, I want to take you through a brief history of how products have evolved.
  44. 3:09 Specifically, I think there are three levels of abstractions, I would call it.
  45. 3:14 Let's imagine a bank to make it easier.
  46. 3:17 Let's imagine a bank.
  47. 3:19 The banks used to have a physical presence before.
  48. 3:21 You would go to a bank branch, you would take the money out by talking to an employee of the bank.
  49. 3:26 And the point here was the person who is doing the task is the employee of a particular company.
  50. 3:32 So you as a user would go in, you would delegate a task to an employee, they would do the task for you and then come back with the result.

Open at this second