vidtheque.

read-only demo

Videos kfSDc2eVLo4

How to Leverage Domain Expertise — Chris Lovejoy, Notius Labs

index_state ready data_status ok

AI Engineer· published 2026-05-16· 0:24:45· en-US· indexed 2026-08-10 19:55

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:05, 1 of 1 keyframes kept
  2. Shot 1, 0:05 to 0:09, 1 of 1 keyframes kept
  3. Shot 2, 0:09 to 0:14, 1 of 1 keyframes kept
  4. Shot 3, 0:14 to 0:29, 1 of 1 keyframes kept
  5. Shot 4, 0:29 to 0:33, 1 of 1 keyframes kept
  6. Shot 5, 0:33 to 1:00, 1 of 1 keyframes kept
  7. Shot 6, 1:00 to 1:28, 1 of 1 keyframes kept
  8. Shot 7, 1:28 to 2:00, 1 of 1 keyframes kept
  9. Shot 8, 2:00 to 2:22, 1 of 1 keyframes kept
  10. Shot 9, 2:22 to 2:23, 1 of 1 keyframes kept
  11. Shot 10, 2:23 to 2:41, 1 of 1 keyframes kept
  12. Shot 11, 2:41 to 3:02, 1 of 1 keyframes kept
  13. Shot 12, 3:02 to 3:32, 1 of 1 keyframes kept
  14. Shot 13, 3:32 to 4:02, 1 of 1 keyframes kept
  15. Shot 14, 4:02 to 4:10, 1 of 1 keyframes kept
  16. Shot 15, 4:10 to 4:25, 1 of 1 keyframes kept
  17. Shot 16, 4:25 to 4:42, 1 of 1 keyframes kept
  18. Shot 17, 4:42 to 4:45, 1 of 1 keyframes kept
  19. Shot 18, 4:45 to 5:02, 1 of 1 keyframes kept
  20. Shot 19, 5:02 to 5:40, 1 of 1 keyframes kept
  21. Shot 20, 5:40 to 5:49, 1 of 1 keyframes kept
  22. Shot 21, 5:49 to 5:52, 1 of 1 keyframes kept
  23. Shot 22, 5:52 to 6:36, 1 of 1 keyframes kept
  24. Shot 23, 6:36 to 7:01, 1 of 1 keyframes kept
  25. Shot 24, 7:01 to 7:26, 0 of 1 keyframes kept
  26. Shot 25, 7:26 to 7:51, 1 of 1 keyframes kept
  27. Shot 26, 7:51 to 8:16, 0 of 1 keyframes kept
  28. Shot 27, 8:16 to 8:44, 1 of 1 keyframes kept
  29. Shot 28, 8:44 to 8:50, 0 of 1 keyframes kept
  30. Shot 29, 8:50 to 8:58, 1 of 1 keyframes kept
  31. Shot 30, 8:58 to 9:32, 1 of 1 keyframes kept
  32. Shot 31, 9:32 to 10:07, 1 of 1 keyframes kept
  33. Shot 32, 10:07 to 10:46, 1 of 1 keyframes kept
  34. Shot 33, 10:46 to 10:49, 1 of 1 keyframes kept
  35. Shot 34, 10:49 to 11:24, 1 of 1 keyframes kept
  36. Shot 35, 11:24 to 11:32, 1 of 1 keyframes kept
  37. Shot 36, 11:32 to 11:42, 0 of 1 keyframes kept
  38. Shot 37, 11:42 to 12:24, 1 of 1 keyframes kept
  39. Shot 38, 12:24 to 12:57, 1 of 1 keyframes kept
  40. Shot 39, 12:57 to 13:30, 1 of 1 keyframes kept
  41. Shot 40, 13:30 to 13:36, 1 of 1 keyframes kept
  42. Shot 41, 13:36 to 14:04, 1 of 1 keyframes kept
  43. Shot 42, 14:04 to 14:10, 1 of 1 keyframes kept
  44. Shot 43, 14:10 to 14:40, 1 of 1 keyframes kept
  45. Shot 44, 14:40 to 15:11, 1 of 1 keyframes kept
  46. Shot 45, 15:11 to 15:42, 1 of 1 keyframes kept
  47. Shot 46, 15:42 to 16:12, 0 of 1 keyframes kept
  48. Shot 47, 16:12 to 16:59, 1 of 1 keyframes kept
  49. Shot 48, 16:59 to 17:05, 0 of 1 keyframes kept
  50. Shot 49, 17:05 to 17:16, 1 of 1 keyframes kept
  51. Shot 50, 17:16 to 17:48, 1 of 1 keyframes kept
  52. Shot 51, 17:48 to 18:22, 1 of 1 keyframes kept
  53. Shot 52, 18:22 to 18:56, 0 of 1 keyframes kept
  54. Shot 53, 18:56 to 19:24, 1 of 1 keyframes kept
  55. Shot 54, 19:24 to 19:34, 0 of 1 keyframes kept
  56. Shot 55, 19:34 to 20:06, 1 of 1 keyframes kept
  57. Shot 56, 20:06 to 20:38, 1 of 1 keyframes kept
  58. Shot 57, 20:38 to 21:13, 1 of 1 keyframes kept
  59. Shot 58, 21:13 to 21:40, 0 of 1 keyframes kept
  60. Shot 59, 21:40 to 22:06, 1 of 1 keyframes kept
  61. Shot 60, 22:06 to 22:33, 0 of 1 keyframes kept
  62. Shot 61, 22:33 to 23:00, 0 of 1 keyframes kept
  63. Shot 62, 23:00 to 23:10, 1 of 1 keyframes kept
  64. Shot 63, 23:10 to 23:23, 1 of 1 keyframes kept
  65. Shot 64, 23:23 to 24:11, 1 of 1 keyframes kept
  66. Shot 65, 24:11 to 24:15, 1 of 1 keyframes kept
  67. Shot 66, 24:15 to 24:30, 1 of 1 keyframes kept
  68. Shot 67, 24:30 to 24:44, 1 of 1 keyframes kept

68 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
285
whisperx 285
chunks
43
from 285 cues
keyframes
57
kept of 68 captured
frames with text
57
1,251 lines read
chapters
0
from the source metadata
keyframe bytes
7.6 MB
word timings on 285 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-10 18:33 2m 02s
stt done — 2026-08-10 18:35 30s
chunk done — 2026-08-10 18:35 0s
text_embed done — 2026-08-10 19:55 0s
keyframe done — 2026-08-10 18:35 2m 12s
ocr done — 2026-08-10 18:37 26s
frame_embed done — 2026-08-10 19:55 10s

Frames, and what the machine read

  • 0:02 #0 done2 line(s)

    shot 0·sharpness 668.7

    1. Al Engineer0.93
    2. EUROPE1.00

  • 0:06 #1 done2 line(s)

    shot 1·sharpness 819.1

    1. PRESENTING SPONSOR0.99
    2. Google DeepMind1.00

  • 0:11 #2 done3 line(s)

    shot 2·sharpness 939.2

    1. PLATINUM SPONSORS0.98
    2. # Braintrust0.95
    3. WorkOS OpenAI0.96

  • 0:16 #3 done13 line(s)

    shot 3·sharpness 664.1

    1. TheD1.00
    2. Native Al Org0.96
    3. How to Le.0.91
    4. Domain Expertise0.99
    5. AlEngineer0.98
    6. Dr Christopher Lovejoy, MD0.99
    7. 200.99
    8. Founder, Notius Labs0.99
    9. EUROPE1.00
    10. 长20.54
    11. AlEngineer1.00
    12. EUROPE1.00
    13. RG1.00

  • 0:32 #4 done17 line(s)

    shot 4·sharpness 2370.7

    1. The Domain-Native Al Organization0.99
    2. sqe70.94
    3. How to Leverage Domain Expertise to Build Better Al Products0.99
    4. snt0.86
    5. Dr Christopher Lovejoy, MD1.00
    6. 2026-04-100.97
    7. Founder, Notius Labs1.00
    8. *★*0.58
    9. AIE1.00
    10. ★1.00
    11. ★1.00
    12. ★1.00
    13. AlEngineer0.98
    14. EUROPE1.00
    15. Google DeepMind1.00
    16. AlEngineer0.98
    17. 20261.00

  • 0:54 #5 done22 line(s)

    shot 5·sharpness 2250.8

    1. medical doctor1.00
    2. Al engineer0.99
    3. Tandem1.00
    4. ***0.71
    5. AIE1.00
    6. 界0.59
    7. UNIVERSITYOF1.00
    8. CAMBRIDGE1.00
    9. ★1.00
    10. aAnterior0.99
    11. ★1.00
    12. ★1.00
    13. NHS1.00
    14. Cera1.00
    15. UCL1.00
    16. zoe0.90
    17. Notius1.00
    18. AlEngineer0.98
    19. EUROPE1.00
    20. Engineering the future of Al1.00
    21. AlEngineer0.97
    22. 20261.00

  • 1:06 #6 done23 line(s)

    shot 6·sharpness 2355.5

    1. medical doctor1.00
    2. Al engineer1.00
    3. OTandem0.97
    4. *★★0.69
    5. AIE1.00
    6. UNIVERSITY OF0.99
    7. CAMBRIDGE1.00
    8. ★1.00
    9. a Anterior0.95
    10. ★1.00
    11. ★1.00
    12. NHS1.00
    13. Cera1.00
    14. 十0.87
    15. UCL1.00
    16. zoe0.90
    17. Notius1.00
    18. AlEngineer0.98
    19. EUROPE1.00
    20. Braintrust1.00
    21. WorkOS OpenAI0.97
    22. AlEngineer0.97
    23. 20260.99

  • 1:32 #7 done52 line(s)

    shot 7·sharpness 3604.0

    1. Make your LLM app a Domain Expert:1.00
    2. World's Fair0.99
    3. How to Build an Expert System -...0.96
    4. a Anterior0.93
    5. 84K views · 7 months ago0.96
    6. Chris Lovejoy, MD0.97
    7. Al Engineer0.99
    8. My hot take: The vertcal Al products that will win are not the ones with0.92
    9. the most sophisticated models or techniques - but those with the best1.00
    10. Vertical Al is a multi-trillion-dollar opportunity. But yı0.99
    11. systemn for incorporating domain expertise.0.96
    12. *★★0.71
    13. ★1.00
    14. MAKE YOUR LLM APP A DOMAIN EXPERT:0.99
    15. 15 chapters Introduction | The last mi0.98
    16. better domain context, not better reasoning models.0.99
    17. customer-specific way that a workflowis performed. Thisis selved by0.92
    18. The LLMs in the product need to understand the industry-specific and0.99
    19. AIE1.00
    20. ★1.00
    21. How to Build an LLM-Native0.99
    22. This is the bet we're taking at Anterior - and why we built a system to0.98
    23. translate domain insights into improvements.0.99
    24. ★1.00
    25. ExpertSystem1.00
    26. We call it an adaptive domain intelligence engine and here's how it0.99
    27. ★1.00
    28. ★1.00
    29. ★1.00
    30. 19:181.00
    31. works:1.00
    32. OUR BET FOR VERTICAL AI APPLICATIONS:1.00
    33. the1.00
    34. system for0.99
    35. sophistication1.00
    36. incorporating0.99
    37. ofyour models0.98
    38. domaininsights1.00
    39. and pipelines0.97
    40. 1:41 pm - 30 Jul 2025 -20.4KVlews0.93
    41. O90.57
    42. t0.94
    43. 010.60
    44. 日2280.78
    45. "but how should I build my org to enable this?”0.97
    46. AlEngineer0.99
    47. EUROPE1.00
    48. "but hows0.97
    49. AlEngineer0.96
    50. AlEngineer0.97
    51. EUROPE1.00
    52. 20261.00

  • 2:17 #8 done16 line(s)

    shot 8·sharpness 666.7

    1. Mak1.00
    2. World'sFair0.95
    3. How1.00
    4. aAnterior0.94
    5. 84Kv0.98
    6. 回A0.66
    7. AlEngineer0.99
    8. MAKE YOUR LLM APP A DOMAIN EXPERT:0.99
    9. EUROPE1.00
    10. How to Build an LLM-Native0.97
    11. ExpertSystem1.00
    12. 19:180.95
    13. "but how should I build m0.97
    14. AlEngineer0.98
    15. EUROPE1.00
    16. ERG1.00

  • 2:22 #9 done53 line(s)

    shot 9·sharpness 3578.9

    1. Make your LLM app a Domain Expert:1.00
    2. World's Fair0.99
    3. How to Build an Expert System -...0.96
    4. a Anterior0.92
    5. 84K views · 7 months ago0.96
    6. Chris Lovejoy, MD0.97
    7. Al Engineer0.98
    8. My hot take: The vertical Al products that will win are not the ones with0.92
    9. the most sophisticated models or techniques - but those with the best0.99
    10. Vertical Al is a multi-trillion-dollar opportunity. But y1.00
    11. system for incorporating domain expertise.0.99
    12. *★*0.54
    13. AIE1.00
    14. ★1.00
    15. ★0.99
    16. How to Build an LLM-Native1.00
    17. MAKE YOUR LLM APP A DOMAIN EXPERT:0.99
    18. 15 chapters Introduction | The last mi0.99
    19. better domain context, not better reasoning models.0.99
    20. customer-specific way that a workflowis performed. Thisis selved by0.92
    21. This is the bet we're taking at Anterior - and why we built a system to0.98
    22. The LLMs in the product need to understand the industry-specific and0.99
    23. translate domain insights into improvements.0.99
    24. ★1.00
    25. ExpertSystem1.00
    26. We call it an adaptive domain intelligence engine and here's how it0.99
    27. ★1.00
    28. ★1.00
    29. ★1.00
    30. 19:181.00
    31. works:1.00
    32. OUR BET FOR VERTICAL AI APPLICATIONS:1.00
    33. the1.00
    34. system for0.99
    35. sophistication1.00
    36. incorporating1.00
    37. ofyour models0.98
    38. domaininsights1.00
    39. and pipelines0.96
    40. 1:41 pm - 30 Jul 2025 -20.4KVlews0.95
    41. O80.55
    42. t0.89
    43. 10.56
    44. 2280.90
    45. 41.00
    46. "but how should I build my org to enable this?”0.98
    47. AlEngineer0.98
    48. EUROPE1.00
    49. "but hows0.98
    50. AlEngineer0.96
    51. AlEngineer0.99
    52. EUROPE1.00
    53. 20260.99

  • 2:39 #10 done15 line(s)

    shot 10·sharpness 1968.2

    1. MY BET FOR VERTICAL AI APPLICATIONS:1.00
    2. *★★0.63
    3. AIE1.00
    4. winning in vertical Al is a0.97
    5. ★1.00
    6. ★1.00
    7. ★1.00
    8. modeiproblem.1.00
    9. organizational1.00
    10. 41.00
    11. AlEngineer0.98
    12. EUROPE1.00
    13. Engineering the future of Al1.00
    14. AlEngineer0.99
    15. 20261.00

  • 2:43 #11 done21 line(s)

    shot 11·sharpness 1568.2

    1. *★*0.58
    2. AIE1.00
    3. ★1.00
    4. ★1.00
    5. ★1.00
    6. Oracle1.00
    7. Evaluator1.00
    8. Architect1.00
    9. Directly add0.98
    10. Define and0.98
    11. Build self-improving1.00
    12. domain expertise1.00
    13. measure quality1.00
    14. systems0.98
    15. 41.00
    16. AlEngineer0.99
    17. EUROPE1.00
    18. Oracle1.00
    19. Engineering the future of Al0.99
    20. AlEngineer0.99
    21. 20260.98

  • 3:15 #12 done40 line(s)

    shot 12·sharpness 2516.4

    1. NEA1.00
    2. Blog0.99
    3. Tomorrow's Titans:0.99
    4. Vertical AI0.98
    5. *★*0.70
    6. ★1.00
    7. by Tiffany Luck and James Kaplan1.00
    8. AI0.82
    9. AIE1.00
    10. ★1.00
    11. Harvey Hits $11B as0.99
    12. ★1.00
    13. ★1.00
    14. ★1.00
    15. ★1.00
    16. Sequoia Triples Down on1.00
    17. Vertical Al Inflection0.99
    18. Bessemer1.00
    19. Partners1.00
    20. Venture1.00
    21. Legal Al startup's valuation surge signals enterprise0.99
    22. infrastructure in professional services.1.00
    23. Al crossing from experimentation to mission-critical0.99
    24. Trend 4: Vertical AI shows potential to dwarf1.00
    25. legacy SaaS with new applications and0.98
    26. business models1.00
    27. Vertical SaaS proved to be a sleeping giant that transformed industries0.99
    28. during the first cloud revolution. Today, the top 20 US publicly traded1.00
    29. vertical SaaS companies represent a combined market capitalization of0.99
    30. -$300 billion, with more than half of these companies having IPO'd in the1.00
    31. last ten years.1.00
    32. AlEngineer0.98
    33. The US Bureau of Labor Statistics cites the Business and Professional0.99
    34. EUROPE1.00
    35. Services industry at 13% of US GDP making this sector alone,0.99
    36. dominated by repetitive language tasks, -10x the size of the software0.99
    37. industry.1.00
    38. Engineering the future of Al1.00
    39. AlEngineer0.98
    40. 20260.99

  • 3:36 #13 done12 line(s)

    shot 13·sharpness 2258.0

    1. ~50% of generative Al0.98
    2. AIE1.00
    3. projects were0.97
    4. abandonedin 20250.99
    5. 40.82
    6. AlEngineer0.94
    7. EUROPE1.00
    8. Source: Gartner (https://www.gartner.com/en/articles/genai-project-failure)0.99
    9. AlEngineer0.97
    10. AlEngineer0.98
    11. EUROPE1.00
    12. 20261.00

  • 4:06 #14 done17 line(s)

    shot 14·sharpness 1521.3

    1. The frontier models are good1.00
    2. enough.1.00
    3. *★*0.65
    4. AIE1.00
    5. ★1.00
    6. ★1.00
    7. ★1.00
    8. The gap is how organizations1.00
    9. operationalize expert0.99
    10. judgment around them.0.98
    11. 40.73
    12. AlEngineer0.98
    13. EUROPE1.00
    14. AlEngineer0.96
    15. AlEngineer1.00
    16. EUROPE1.00
    17. 20261.00

  • 4:23 #15 done24 line(s)

    shot 15·sharpness 3255.8

    1. The most common mistakes I see1.00
    2. 1 . Not hiring domain experts (or0.99
    3. hiring them too late)1.00
    4. *★*0.67
    5. AIE1.00
    6. ★1.00
    7. ★1.00
    8. 2. Hiring the wrong kind of domain0.97
    9. ★1.00
    10. ★1.00
    11. ★1.00
    12. expert1.00
    13. 3. Not fitting them into your org, not0.98
    14. leveraging them appropriately1.00
    15. 40.99
    16. AlEngineer0.99
    17. EUROPE1.00
    18. expert1.00
    19. 3. Not fitting0.99
    20. leveraging1.00
    21. AlEngineer0.96
    22. AlEngineer0.99
    23. EUROPE1.00
    24. 20261.00

  • 4:37 #16 done27 line(s)

    shot 16·sharpness 5212.0

    1. The most common mistakes I see0.99
    2. 1. Do I really need a0.98
    3. domain expert(s)?1.00
    4. 1. Not hiring domain experts (or0.98
    5. hiring them too late)1.00
    6. *★*0.63
    7. AIE1.00
    8. ★1.00
    9. ★1.00
    10. ★1.00
    11. 2. Hiring the wrong kind of domain0.98
    12. 2. Who do I0.98
    13. expert1.00
    14. need?1.00
    15. 3. Not fitting them into your org, not0.98
    16. 3. How should I0.97
    17. leveraging them appropriately1.00
    18. leverage them?1.00
    19. 40.76
    20. AlEngineer0.98
    21. The Oracle - Evaluator- Architect Framework0.97
    22. EUROPE1.00
    23. 3. Not fitting0.95
    24. leveraging1.00
    25. Engineering the future of Al1.00
    26. AlEngineer1.00
    27. 20261.00

  • 4:44 #17 done13 line(s)

    shot 17·sharpness 1186.2

    1. AIE1.00
    2. 1. Do I really need a domain expert?0.99
    3. ★0.97
    4. ★1.00
    5. ★1.00
    6. ★1.00
    7. 41.00
    8. AlEngineer0.99
    9. EUROPE1.00
    10. Engineering the future of Al0.99
    11. AlEngineer0.99
    12. EUROPE1.00
    13. 20261.00

  • 4:48 #18 done17 line(s)

    shot 18·sharpness 1528.2

    1. Judgment1.00
    2. Appraising Al0.97
    3. AIE1.00
    4. requires1.00
    5. quality requires1.00
    6. ★1.00
    7. domain1.00
    8. judgment.1.00
    9. expertise.1.00
    10. 40.92
    11. AlEngineer0.97
    12. EUROPE1.00
    13. jud1.00
    14. Engineering the future of Al1.00
    15. AlEngineer1.00
    16. 20261.00
    17. EUROPE1.00

  • 5:29 #19 done7 line(s)

    shot 19·sharpness 585.7

    1. AlEngineer0.98
    2. Appraising Al0.99
    3. EUROPE1.00
    4. quality requires1.00
    5. judgment.1.00
    6. AlEngineer0.98
    7. EUROPE1.00

  • 5:46 #20 done16 line(s)

    shot 20·sharpness 1514.7

    1. Judgment1.00
    2. Appraising Al0.96
    3. AIE1.00
    4. requires1.00
    5. quality requires1.00
    6. domain1.00
    7. judgment.1.00
    8. expertise.1.00
    9. 40.96
    10. AlEngineer0.95
    11. EUROPE1.00
    12. judg0.86
    13. AlEngineer0.98
    14. AlEngineer0.97
    15. EUROPE1.00
    16. 20261.00

  • 5:49 #21 done12 line(s)

    shot 21·sharpness 1036.1

    1. AIE1.00
    2. 2. Who do I need?0.99
    3. ★1.00
    4. ★1.00
    5. ★1.00
    6. 40.97
    7. AlEngineer0.96
    8. EUROPE1.00
    9. AlEngineer0.97
    10. AlEngineer0.97
    11. EUROPE1.00
    12. 20261.00

  • 6:30 #22 done21 line(s)

    shot 22·sharpness 1573.4

    1. AIE1.00
    2. ★1.00
    3. ★1.00
    4. ★1.00
    5. ★1.00
    6. Oracle1.00
    7. Evaluator1.00
    8. Architect1.00
    9. Directly add0.99
    10. Define and0.98
    11. Build self-improving1.00
    12. domain expertise1.00
    13. measure quality0.97
    14. systems1.00
    15. 40.99
    16. AlEngineer0.98
    17. EUROPE1.00
    18. Oracle1.00
    19. Engineering the future of Al0.99
    20. AlEngineer0.99
    21. 20261.00

  • 6:58 #23 done18 line(s)

    shot 23·sharpness 1697.5

    1. AIE1.00
    2. Oracle1.00
    3. ★1.00
    4. ★1.00
    5. ★1.00
    6. assess1.00
    7. improve1.00
    8. Domain1.00
    9. Domain1.00
    10. expert1.00
    11. expert1.00
    12. 40.92
    13. AlEngineer0.98
    14. EUROPE1.00
    15. Engineering the future of Al1.00
    16. AlEngineer0.99
    17. EUROPE1.00
    18. 20261.00

Transcript

285 cues· 4,663 words· 26,225 chars

  1. 0:15 Okay, so welcome everybody.
  2. 0:17 Hi, my name is Christopher Lovejoy, and I'm gonna talk about how to leverage domain expertise to build better AI products.
  3. 0:26 And the way I believe you can do this is by building what I call a domain-native AI organization.
  4. 0:31 So I'm gonna talk about what that looks like.
  5. 0:34 A brief background about me, because this is relevant.
  6. 0:37 So I started out my career as a medical doctor.
  7. 0:39 I trained in the University of Cambridge and then worked in the NHS for several years.
  8. 0:44 And I then, in 2018, moved into AI space, training and building models, and working in various organizations, including Tandem, which is the largest AI product provider in the UK in terms of adoption.
  9. 0:59 Also Anterior, I was the first employee and it's a square back startup forming prior authorization in the US and then also various other startups as well.
  10. 1:11 And my kind of challenge at all of these different companies was we're building some kind of product that bakes in domain expertise.
  11. 1:19 How can you do that?
  12. 1:21 How can you leverage that in a way that then builds a differentiated AI product?
  13. 1:30 So I talked a bit about this at the last AI engineer conference in San Francisco.
  14. 1:34 I shared my thesis, which is that the system for incorporating domain insights is more important than the sophistication of your models, of your pipelines.
  15. 1:43 I talked about the last mile problem, which is this challenge of getting your product to really understand the specific nuances of the workflows of the use cases of your customer that you're serving.
  16. 1:55 And this talk was,
  17. 1:57 seen by a lot of people, about 100,000 people saw this across different platforms.
  18. 2:02 Many of them reached out to me, and the kind of most common question that they had was, okay, but how do I build my organization?
  19. 2:09 Like, I'm kind of on board.
  20. 2:10 I get that domain expertise is important, but how do I build my organization to actually enable this to, like, what kind of domain expertise should I hire?
  21. 2:18 Where should I put them in my organization to enable this to take place?
  22. 2:20 So that's what I'm gonna talk about today.
  23. 2:24 And I've heard people say that winning in vertical AI, you kind of want to get the best model.
  24. 2:30 And actually, I don't think this is true.
  25. 2:31 I think that fundamentally, winning in vertical AI is an organizational problem.
  26. 2:44 So I have this framework based on different organizations I've seen and worked at and how they're baking domain expertise.
  27. 2:53 which is that you can do it in three main ways.
  28. 2:55 You can have your domain expert as an oracle, as an evaluator, as an architect, and I'm gonna talk a bit more about those in more detail.
  29. 3:04 But just stepping back, why do we care about vertical AI?
  30. 3:08 Ultimately, it comes down to the fact that vertical AI is a big opportunity.
  31. 3:11 So a lot of VCs are talking about, you know, this is the next big thing, as a lot of startups raising big amounts of money on this potential, this promise.
  32. 3:20 And as Bessma pointed out, vertical AI, you know, we had vertical SaaS, and that was a 50 billion or something dollar market.
  33. 3:26 But actually, now AI is moving into the kind of labor force.
  34. 3:29 And that's like a multi trillion dollar market.
  35. 3:33 but we've not yet really seen success at scale.
  36. 3:37 And according to Gartner, about 50% of all generative AI projects were abandoned last year.
  37. 3:42 And I think there's many reasons for this, but my take is that one of the core reasons for this is that we're often building AI products and AI systems without really having a deep kind of understanding of exactly what workflows we're automating and exactly how the domain experts would perform these kind of processes.
  38. 4:04 So just to reiterate, my belief is front-end models are good enough, but the gap is now how do organizations operationalize the expert judgment around them?
  39. 4:11 And the three most common mistakes that I have seen are, firstly, not hiring domain experts, or hiring them too late, secondly, hiring the wrong kind of domain experts, and then finally, not fitting them into your organization appropriately, not leveraging them correctly.
  40. 4:25 And so this maps to three questions, which I'm gonna address in this talk, which is, do you really need a domain expert?
  41. 4:31 What's the why?
  42. 4:33 If so, who do you need?
  43. 4:34 And then, how do you leverage them?
  44. 4:36 So I touched on the first one, and then this Oracle evaluator architect framework I'm gonna go through answers those second two.
  45. 4:43 So, do you really need a domain expert?
  46. 4:46 My take is that the answer is yes, and it's because
  47. 4:51 Appraising AI quality is something that's very important to do in your company.
  48. 4:55 You want to be able to make decisions between different approaches based on the kind of output that they give.
  49. 5:00 And your company needs to have a sense of what good AI quality looks like.
  50. 5:04 And that ultimately requires judgment.

Open at this second