vidtheque.

read-only demo

Videos B9h9ovW5H9U

Why your agents need decision traces, not just documents — Zach Blumenfeld, Neo4j

index_state ready data_status ok

AI Engineer· published 2026-05-29· 0:20:12· en-US· indexed 2026-08-10 23:06

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 0:45, 1 of 1 keyframes kept
  5. Shot 4, 0:45 to 1:16, 0 of 1 keyframes kept
  6. Shot 5, 1:16 to 1:35, 1 of 1 keyframes kept
  7. Shot 6, 1:35 to 2:09, 1 of 1 keyframes kept
  8. Shot 7, 2:09 to 2:38, 1 of 1 keyframes kept
  9. Shot 8, 2:38 to 3:08, 0 of 1 keyframes kept
  10. Shot 9, 3:08 to 3:32, 1 of 1 keyframes kept
  11. Shot 10, 3:32 to 3:58, 1 of 1 keyframes kept
  12. Shot 11, 3:58 to 4:44, 1 of 1 keyframes kept
  13. Shot 12, 4:44 to 4:45, 1 of 1 keyframes kept
  14. Shot 13, 4:45 to 5:13, 1 of 1 keyframes kept
  15. Shot 14, 5:13 to 5:15, 1 of 1 keyframes kept
  16. Shot 15, 5:15 to 5:55, 1 of 1 keyframes kept
  17. Shot 16, 5:55 to 6:42, 1 of 1 keyframes kept
  18. Shot 17, 6:42 to 6:45, 0 of 1 keyframes kept
  19. Shot 18, 6:45 to 6:50, 0 of 1 keyframes kept
  20. Shot 19, 6:50 to 6:51, 1 of 1 keyframes kept
  21. Shot 20, 6:51 to 7:21, 1 of 1 keyframes kept
  22. Shot 21, 7:21 to 7:46, 1 of 1 keyframes kept
  23. Shot 22, 7:46 to 8:11, 0 of 1 keyframes kept
  24. Shot 23, 8:11 to 8:33, 1 of 1 keyframes kept
  25. Shot 24, 8:33 to 8:56, 1 of 1 keyframes kept
  26. Shot 25, 8:56 to 9:24, 1 of 1 keyframes kept
  27. Shot 26, 9:24 to 9:27, 1 of 1 keyframes kept
  28. Shot 27, 9:27 to 10:06, 1 of 1 keyframes kept
  29. Shot 28, 10:06 to 10:09, 1 of 1 keyframes kept
  30. Shot 29, 10:09 to 10:13, 1 of 1 keyframes kept
  31. Shot 30, 10:13 to 10:15, 1 of 1 keyframes kept
  32. Shot 31, 10:15 to 10:17, 1 of 1 keyframes kept
  33. Shot 32, 10:17 to 10:45, 1 of 1 keyframes kept
  34. Shot 33, 10:45 to 11:12, 0 of 1 keyframes kept
  35. Shot 34, 11:12 to 11:39, 1 of 1 keyframes kept
  36. Shot 35, 11:39 to 11:48, 0 of 1 keyframes kept
  37. Shot 36, 11:48 to 12:04, 1 of 1 keyframes kept
  38. Shot 37, 12:04 to 12:07, 0 of 1 keyframes kept
  39. Shot 38, 12:07 to 12:36, 1 of 1 keyframes kept
  40. Shot 39, 12:36 to 13:05, 0 of 1 keyframes kept
  41. Shot 40, 13:05 to 13:07, 1 of 1 keyframes kept
  42. Shot 41, 13:07 to 13:47, 1 of 1 keyframes kept
  43. Shot 42, 13:47 to 14:14, 1 of 1 keyframes kept
  44. Shot 43, 14:14 to 14:40, 0 of 1 keyframes kept
  45. Shot 44, 14:40 to 14:56, 1 of 1 keyframes kept
  46. Shot 45, 14:56 to 15:24, 1 of 1 keyframes kept
  47. Shot 46, 15:24 to 15:51, 0 of 1 keyframes kept
  48. Shot 47, 15:51 to 16:18, 1 of 1 keyframes kept
  49. Shot 48, 16:18 to 16:45, 0 of 1 keyframes kept
  50. Shot 49, 16:45 to 17:13, 1 of 1 keyframes kept
  51. Shot 50, 17:13 to 17:40, 1 of 1 keyframes kept
  52. Shot 51, 17:40 to 18:07, 1 of 1 keyframes kept
  53. Shot 52, 18:07 to 18:35, 0 of 1 keyframes kept
  54. Shot 53, 18:35 to 19:02, 1 of 1 keyframes kept
  55. Shot 54, 19:02 to 19:30, 0 of 1 keyframes kept
  56. Shot 55, 19:30 to 19:57, 1 of 1 keyframes kept
  57. Shot 56, 19:57 to 20:11, 1 of 1 keyframes kept
  58. Shot 57, 20:11 to 20:12, 0 of 1 keyframes kept

58 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
159
whisperx 159
chunks
33
from 159 cues
keyframes
43
kept of 58 captured
frames with text
43
1,687 lines read
chapters
0
from the source metadata
keyframe bytes
7.5 MB
word timings on 159 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-10 23:01 1m 27s
stt done — 2026-08-10 23:03 26s
chunk done — 2026-08-10 23:03 0s
text_embed done — 2026-08-10 23:03 9s
keyframe done — 2026-08-10 23:03 1m 48s
ocr done — 2026-08-10 23:05 32s
frame_embed done — 2026-08-10 23:06 8s

Frames, and what the machine read

  • 0:03 #0 done2 line(s)

    shot 0·sharpness 658.4

    1. AlEngineer0.98
    2. EUROPE1.00

  • 0:08 #1 done2 line(s)

    shot 1·sharpness 827.9

    1. PRESENTINGSPONSOR1.00
    2. Google DeepMind1.00

  • 0:13 #2 done3 line(s)

    shot 2·sharpness 907.7

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

  • 0:39 #3 done8 line(s)

    shot 3·sharpness 2014.1

    1. neo4j0.99
    2. Context Graphs 1010.99
    3. AIE1.00
    4. ★1.00
    5. What They Are and How to Get Started in0.98
    6. 5 Minutes1.00
    7. Context Graphs 1010.99
    8. Engineering the future of Al0.98

  • 0:49 #4 skipped

    shot 4·duplicate of #3

  • 1:22 #5 done34 line(s)

    shot 5·sharpness 3079.5

    1. Context Graphs1.00
    2. Content Geaph0.87
    3. foundation1.00
    4. capital1.00
    5. Al's trillion-dollar opportunity:1.00
    6. Context graphs1.00
    7. **0.85
    8. ★1.00
    9. AIE1.00
    10. https://foundationcapital.com/context-graphs-ais-trillion-dollar-opportunity/1.00
    11. ★1.00
    12. ★1.00
    13. ★1.00
    14. ★1.00
    15. CONTEXT GRAPHS: CAPTURING THE 'WHY'0.99
    16. ENTERNRIM0.86
    17. Context1.00
    18. Graphs1.00
    19. https://arxiv.org/abs/2406.111600.98
    20. €dharmesh0.93
    21. 18:521.00
    22. Hands On With Context Graphs And0.99
    23. In The Age of Al0.99
    24. Context Graphs: Capturing The "Why"0.98
    25. Oharmesh Shah0.97
    26. Sudhir Hasbe (Ep 43)1.00
    27. Graphs, Graphs Everywhere-with1.00
    28. Neo4j1.00
    29. Building Al systems that don't just make decisions, but remember and1.00
    30. use the "why".1.00
    31. Willa Lyon 9 min nead Jan 14, 20260.87
    32. 21.00
    33. AlEngineer0.97
    34. EUROPE1.00

  • 2:01 #6 done15 line(s)

    shot 6·sharpness 1899.3

    1. What Do Agents Need to be Accurate?1.00
    2. Knowledge Base1.00
    3. Context Graph1.00
    4. Information required to answer0.99
    5. Information required to make1.00
    6. *★★0.60
    7. questions1.00
    8. decisions1.00
    9. AIE1.00
    10. ★1.00
    11. ★1.00
    12. ★1.00
    13. ★1.00
    14. AlEngineer0.96
    15. EUROPE1.00

  • 2:35 #7 done26 line(s)

    shot 7·sharpness 4049.0

    1. What Do Agents Need to be Accurate?1.00
    2. Knowledge Base1.00
    3. Context Graph1.00
    4. Information required to answer0.99
    5. Information required to make0.99
    6. ***0.60
    7. ★0.99
    8. questions1.00
    9. decisions1.00
    10. AIE1.00
    11. ★0.98
    12. ★1.00
    13. ★1.00
    14. ★1.00
    15. Ex: Financial Analyst Agent1.00
    16. “Approve credit limit increase for Jessica Norris? Requesting £ 25K”0.98
    17. Retrieve:1.00
    18. Customer info, transactions, and policies1.00
    19. Response:1.00
    20. "Moderate-to-high risk score, elevated-risk1.00
    21. jurisdiction. Per High-Value Transaction policy,1.00
    22. warrants scrutiny. Key factors to investigate :0.99
    23. account activity patterns, transaction velocity, and1.00
    24. KYC status. Recommend a manual review before0.99
    25. proceeding."1.00
    26. Engineering the future of Al1.00

  • 2:59 #8 skipped

    shot 8·duplicate of #7

  • 3:11 #9 done23 line(s)

    shot 9·sharpness 3302.1

    1. What Do Agents Need to be Accurate?1.00
    2. Knowledge Base1.00
    3. Context Graph1.00
    4. Information required to answer0.98
    5. Information required to make1.00
    6. ***0.60
    7. questions1.00
    8. decisions1.00
    9. AIE1.00
    10. ★1.00
    11. ★1.00
    12. ★1.00
    13. ★1.00
    14. Facts, entities, policies, current0.99
    15. Decision traces, precedents,1.00
    16. state1.00
    17. expert-validated outcomes0.98
    18. Enables accurate retrieval. It tells0.98
    19. Enables agent to act with0.99
    20. the agent what is true.0.99
    21. institutional knowledge. It tells the1.00
    22. agent what to do & why.0.99
    23. Engineering the future of Al1.00

  • 3:42 #10 done30 line(s)

    shot 10·sharpness 1952.8

    1. Context Graph Data Model1.00
    2. ENTITIES (What exists)0.99
    3. Financial Services Context Graph0.99
    4. Person, Account, Transaction, Organization0.99
    5. HAS_TXN0.98
    6. Account1.00
    7. *★★0.54
    8. ★1.00
    9. ENTITIES (What exists)0.98
    10. OMNS0.87
    11. AIE1.00
    12. ★1.00
    13. ★1.00
    14. EVENTS (What happened)1.00
    15. Person1.00
    16. ★1.00
    17. ★1.00
    18. ★1.00
    19. DECIDED1.00
    20. Decisions, Transactions, Approvals, Rejections0.99
    21. EXPLOYED_BY0.96
    22. Decision1.00
    23. APPLIED1.00
    24. Policy0.99
    25. DECISIONS (The "why*)0.99
    26. CONTEXT (The why)1.00
    27. Policies applied, Risk factors, Employee reasoning1.00
    28. Blog:1.00
    29. AlEngineer0.97
    30. EUROPE1.00

  • 4:08 #11 done46 line(s)

    shot 11·sharpness 3606.4

    1. Context Graph Demo1.00
    2. Financial Services Agent1.00
    3. context-graph-demo.vercel.app1.00
    4. DATA SOURCES1.00
    5. BACKEND SERVICES0.98
    6. FRONTEND & USER INTERFACE1.00
    7. ★1.00
    8. ★1.00
    9. AIE1.00
    10. ★1.00
    11. ★1.00
    12. Support Ticket1.00
    13. System0.97
    14. Ingestion/ETL1.00
    15. Data1.00
    16. 10 MCP Tools1.00
    17. + Claude Agent SDK0.99
    18. FastAPI Application0.99
    19. Claude Agent1.00
    20. Data Transfer1.00
    21. API Calls &1.00
    22. Application1.00
    23. Next.Js UI0.94
    24. User Actions1.00
    25. & Queries0.96
    26. Salesforce)1.00
    27. CRM (e.g.,0.95
    28. (Embeddings)1.00
    29. OpenAI API0.98
    30. Chakra Ul v30.97
    31. Neo4j Context Graph & GDS0.99
    32. Internal1.00
    33. X0.98
    34. NVL Graphs1.00
    35. Business Data1.00
    36. AuraDS0.97
    37. Neo4j1.00
    38. Search1.00
    39. Vector1.00
    40. Graph Data Science Algorithms0.99
    41. (GDS) -FastRP, KNN, Node0.97
    42. (Visualization)1.00
    43. Similarity, Louvain, PageRank1.00
    44. Prerequisites: Python 3.11+, Node.js 18+, Neo4j AuraDS, Anthropic API Key, OpenAI API Key0.98
    45. AlEngineer0.97
    46. EUROPE1.00

  • 4:44 #12 done50 line(s)

    shot 12·sharpness 2612.9

    1. docs.google.com/presentation/d/1SUohQetFkVVs8OTvi02w8Uglb-WVZgNNtWCLuTQ871l/ed0.99
    2. To exit full screen, press and hold0.99
    3. f_0_1340.89
    4. ☆0.77
    5. Work0.90
    6. File Edit View Insert Format Slide Arrange0.98
    7. Context Graph - 5 min☆0.94
    8. Tools0.99
    9. Extensions1.00
    10. Help1.00
    11. ③0.65
    12. Slideshow1.00
    13. Share0.98
    14. Q. Menus0.90
    15. Tr0.99
    16. Background1.00
    17. Layout1.00
    18. Theme1.00
    19. Transition1.00
    20. A0.91
    21. ★1.00
    22. ★1.00
    23. Context Graph Demo0.98
    24. ★1.00
    25. AIE1.00
    26. Financial Services Agent1.00
    27. ★1.00
    28. ★1.00
    29. context-graph-demo.vercel.app0.99
    30. 回0.76
    31. ★1.00
    32. ★1.00
    33. ★1.00
    34. DATA SOURCES1.00
    35. BACKEND SERVICES1.00
    36. FRONTEND & USER INTERFACE0.98
    37. Support Ticket0.93
    38. System0.94
    39. B0.75
    40. FastAPI Application0.96
    41. « Claude Agent SOK0.89
    42. 口0.98
    43. 690.69
    44. Chakra UI v30.88
    45. ΩNoo4j Contat Graph & GDS0.85
    46. D0.93
    47. Interral0.84
    48. Q0.55
    49. 回0.56
    50. Engineering the future of Al0.99

  • 5:07 #13 done81 line(s)

    shot 13·sharpness 2246.9

    1. C0.56
    2. à context-graph-demo.vercel.app0.97
    3. @☆0.73
    4. Work1.00
    5. Context Graph Demo1.00
    6. Al-powered decision tracing with Neo4j1.00
    7. Blog Post1.00
    8. Issues1.00
    9. GitHub1.00
    10. About & Schema0.97
    11. Al Assistant0.99
    12. Context Graph1.00
    13. Decision Trace1.00
    14. Ask questions about customers, decisions, and1.00
    15. Visualize entities, decisions, and causal relationships0.99
    16. Inspect reasoning, precedents, and causal chains0.99
    17. policies1.00
    18. patterns.1.00
    19. Decisions in Graph1.00
    20. Now let me search for similar credit limit1.00
    21. increase decisions to understand1.00
    22. precedents for this type of request.1.00
    23. Decisions visible in the Context Graph. Click to view1.00
    24. details, or double-click nodes in the graph to expand.1.00
    25. ★1.00
    26. ★1.00
    27. DECISION: REJECT X0.99
    28. 4 decisions in graph0.99
    29. ★1.00
    30. ★1.00
    31. ★1.00
    32. AIE1.00
    33. ★1.00
    34. ★1.00
    35. ★1.00
    36. ★1.00
    37. Requested Amount: $25,000 increase0.99
    38. Final Recommendation: REJECTION1.00
    39. Credit Limit Increase Request -0.98
    40. Jessica Norris1.00
    41. Decision made for compliance case. Type:1.00
    42. review. Standard review completed with...0.99
    43. 10/20/20251.00
    44. review1.00
    45. compliance1.00
    46. 68% conf1.00
    47. Analysis Summary1.00
    48. Based on my comprehensive analysis of0.99
    49. rejection fraud1.00
    50. 79% conf1.00
    51. precedents in the Context Graph, I0.99
    52. Jessica Norris's profile and historical1.00
    53. Transaction blocked. Pattern matches known1.00
    54. fraud typology FT-8291. Velocity check failed: 1...0.99
    55. recommend rejecting this credit limit0.99
    56. 6/27/20251.00
    57. 4 risk factors0.97
    58. increase request due to multiple high-1.00
    59. risk factors.1.00
    60. rejection compliance1.00
    61. 71% conf1.00
    62. Key Risk Factors Identified:1.00
    63. Decision made for compliance case. Type:0.99
    64. Previous Fraud & Compliance1.00
    65. rejection. Standard review completed with...0.98
    66. Rejections1.00
    67. Fraud rejection (June 2025):0.98
    68. 5/21/20251.00
    69. 2 risk factors1.00
    70. Transaction blocked matching1.00
    71. rejection credit0.99
    72. 71% conf1.00
    73. Ask about customers, decisions, or1.00
    74. policies...1.00
    75. expand1.00
    76. Click to inspect0.97
    77. Credit application rejected. Credit score of 7091.00
    78. below minimum threshold of 620. Recent...0.98
    79. 9/13/20241.00
    80. 1 risk factors1.00
    81. Engineering the future of Al1.00

  • 5:15 #14 done48 line(s)

    shot 14·sharpness 2576.9

    1. context-graph-demo.vercel.app1.00
    2. ☆0.96
    3. Work0.97
    4. Context Graph - 5 min ☆0.93
    5. File Edit View Insert Format Slide0.99
    6. Arrange1.00
    7. Tools1.00
    8. Extensions1.00
    9. Help1.00
    10. ③0.55
    11. Slideshow1.00
    12. Share0.98
    13. Q Menus0.86
    14. 宁0.67
    15. @ 92%0.90
    16. Tr0.96
    17. Background1.00
    18. Layout1.00
    19. Theme1.00
    20. Transition1.00
    21. ★1.00
    22. ★1.00
    23. Context Graph Demo1.00
    24. ★0.96
    25. AIE1.00
    26. Financial Services Agent1.00
    27. ★1.00
    28. ★1.00
    29. context-graph-demo.vercel.app1.00
    30. 回0.58
    31. ★1.00
    32. ★1.00
    33. ★1.00
    34. DATA SOURCES1.00
    35. BACKEND SERVICES1.00
    36. FRONTEND & USER INTERFACE0.99
    37. Suppert Ticket0.97
    38. System0.96
    39. FastAPI Application0.97
    40. + Claude Agent SDK0.98
    41. 口0.99
    42. 690.64
    43. Chakna UI v30.86
    44. Neosj Cantxt Oraph & GOS0.84
    45. D0.83
    46. Q0.74
    47. AlEngineer0.96
    48. EUROPE1.00

  • 5:50 #15 done83 line(s)

    shot 15·sharpness 2149.7

    1. 2ΩNe040.72
    2. Conte0.99
    3. Scheo0.87
    4. t context-graph-demo.vercel.app0.96
    5. 我☆0.67
    6. Work0.99
    7. Context Graph Demo0.99
    8. Al-powered decision tracing with Neo4j0.99
    9. Blog Post1.00
    10. Issues0.99
    11. GitHub1.00
    12. About & Schema1.00
    13. Al Assistant0.99
    14. Context Graph1.00
    15. Decision Trace1.00
    16. Ask questions about customers, decisions, and0.99
    17. Visualize entities, decisions, and causal relationships1.00
    18. Inspect reasoning. precedents, and causal chains1.00
    19. policies1.00
    20. Now let me get more details on the0.99
    21. Decisions in Graph1.00
    22. ★1.00
    23. ★1.00
    24. AIE1.00
    25. ★1.00
    26. previous rejection decision and1.00
    27. analyze fraud patterns for her1.00
    28. accounts.1.00
    29. FINAL DECISION: REJECT1.00
    30. Recommendation1.00
    31. Decisions visible in the Context Graph. Click to0.98
    32. expand.1.00
    33. view details, or double-click nodes in the graph to1.00
    34. 5 decisions in graph0.97
    35. ★1.00
    36. ★1.00
    37. ★1.00
    38. ★1.00
    39. I recommend REJECTING the1.00
    40. $25,000 credit limit increase1.00
    41. request for Jessica Norris.1.00
    42. Decision made for compliance case. Type:0.99
    43. review. Standard review completed with...0.99
    44. review1.00
    45. compliance0.96
    46. 68% conf1.00
    47. Key Risk Factors1.00
    48. 10/20/20251.00
    49. Prior Fraud Rejection (June 2025)0.99
    50. - Critical Issue0.98
    51. Transaction blocked matching0.98
    52. known fraud typology FT-82910.99
    53. Velocity check failed; 140.98
    54. fraud typology FT-8291. Velocity check failed..0.97
    55. Transaction blocked. Pattern matches known1.00
    56. rejection fraud0.99
    57. 79% conf0.99
    58. transactions in 29 minutes1.00
    59. 6/27/20251.00
    60. 4 risk factors1.00
    61. Geographic anomaly: IP location1.00
    62. inconsistent with account1.00
    63. rejection credit0.99
    64. 88% conf0.97
    65. address1.00
    66. Credit limit increase of $25,000 rejected for1.00
    67. after-hours activity1.00
    68. Multiple red flags: new device,1.00
    69. high amount, round amount,1.00
    70. Jessica Norris. Customer has significant...1.00
    71. 4/9/20261.00
    72. Multiple Compliance Violations0.99
    73. Two separate compliance1.00
    74. Decision made for compliance case. Type:1.00
    75. rejection compliance0.99
    76. 71% conf0.99
    77. rejection. Standard review completed with...1.00
    78. Ask about customers, decisions, or1.00
    79. policies..0.98
    80. 5/21/20251.00
    81. 2 risk factors1.00
    82. Al Engineer0.91
    83. EUROPE1.00

  • 6:37 #16 done96 line(s)

    shot 16·sharpness 2862.8

    1. Noo40.75
    2. Schec0.94
    3. workspace.neo4j.io/workspace/query1.00
    4. 8☆0.80
    5. Work0.95
    6. jeo4j0.68
    7. Explore1.00
    8. Query1.00
    9. Import1.00
    10. 613240c5.databases.neo4j.io /neo4j0.98
    11. Send feedback0.99
    12. e0.93
    13. Database information1.00
    14. Nodes (3.536)0.96
    15. Account1.00
    16. Alert1.00
    17. Started stresming 25 reconds after 13 ms and completed after 14 m0.96
    18. Employee1.00
    19. Message0.99
    20. MATCH p=(:Decision)-[:CAUSED|INFLUENCED[PRECEDENT_FOR]->(:Dec1s1on) RETURN p LIMIT 10e0.97
    21. ()0.70
    22. ReasoningStep1.00
    23. SupportTicket1.00
    24. Organization0.98
    25. Tool1.00
    26. ReasoningTrace0.98
    27. ToolCall1.00
    28. Preference1.00
    29. Graph1.00
    30. Table0.98
    31. traud0.95
    32. RAW1.00
    33. CAUBED0.96
    34. community_id0.97
    35. confidence_scor 0.9540.99
    36. 28691.00
    37. 00.69
    38. 00.60
    39. ★1.00
    40. ★1.00
    41. Show all nodes (I more)0.98
    42. fraud1.00
    43. created_at1.00
    44. 2026-04-08T14:14:55.244001.00
    45. ★1.00
    46. AIE1.00
    47. ★1.00
    48. Relationships (13,463)0.99
    49. decision_:timesta 2026-2-1121:52:0284213900.85
    50. 0000Z0.99
    51. ★1.00
    52. ★1.00
    53. ABOUT APPLIED_POLICY0.99
    54. mp1.00
    55. 000Z0.99
    56. ★1.00
    57. ★1.00
    58. ★1.00
    59. ASSIGNED_TO1.00
    60. BELONGS_TO_ACCOUNT_COMMUNITY1.00
    61. support0.89
    62. fast_rp_embeddi [.0.010616122744977474,.0.000.92
    63. decision_type1.00
    64. "approval"0.97
    65. 00.94
    66. CAUSED COUNTERPARTY1.00
    67. FROM_ACCOUNT HAS_SIMILAR_FACTORS0.99
    68. BELONGS_TO_DECISION_COMMUNITY1.00
    69. ng1.00
    70. 668194413185, 0.1539906260.97
    71. 25265680626034737,-0.2180.98
    72. 3954119682312.-0.00432630.99
    73. 33514404, 0.12980301678180.97
    74. INFLUENCED MADE_BY OWNS0.98
    75. 0695, -0.19989818334579...0.94
    76. OWNS_ACCOUNT PRECEDENT_FOR0.99
    77. REGARDING RELATED_TO RESOLVED_BY0.97
    78. SHARES_HIGH_PERCENTAGE_OF_TRANSACTI...0.99
    79. SUBMITTED_BY1.00
    80. support0.96
    81. tradinn0.94
    82. 4· Q x0.50
    83. id1.00
    84. reasoning1.00
    85. b5647123-2574-4ab7-bf3e-0.95
    86. d7b63ae6cc880.94
    87. "Decision made for support c0.99
    88. Show all1.00
    89. Show all relationships (3 more)0.98
    90. Property keys1.00
    91. :welcome0.99
    92. 口0.88
    93. GUIDE1.00
    94. DATASET1.00
    95. Last update: 10:47:20 AM0.85
    96. Engineering the future of Al1.00

  • 6:44 #17 skipped

    shot 17·duplicate of #14

  • 6:49 #18 skipped

    shot 18·duplicate of #11

  • 6:51 #19 done6 line(s)

    shot 19·sharpness 807.4

    1. AIE1.00
    2. ★1.00
    3. ★1.00
    4. ★1.00
    5. ★1.00
    6. Engineering the future of Al1.00

  • 7:03 #20 done47 line(s)

    shot 20·sharpness 3313.7

    1. How the Agent Traces Decision History1.00
    2. Tool Calls (7)1.00
    3. View tool parameters0.99
    4. and results1.00
    5. Tool Call search_customer1.00
    6. REJECT or ESCALATE — grounded in institutional knowledge, not just a credit score1.00
    7. Arguments:1.00
    8. "query": "Jessica0.99
    9. 11.00
    10. Query Context Graph1.00
    11. Norris",0.95
    12. "limit": 50.98
    13. Agent retrieves Jessica's full profile: accounts, transactions, employer, risk tier1.00
    14. Tool Call find_precedents1.00
    15. Arguments:1.00
    16. ***0.66
    17. AIE1.00
    18. 21.00
    19. Trace Decision History1.00
    20. increase $25000",0.97
    21. "category": "credit",1.00
    22. "linit": 100.94
    23. "scenario": "credit limit0.99
    24. ★1.00
    25. ★1.00
    26. ★1.00
    27. Past fraud flag (April 2025), compliance rejections, velocity check failures1.00
    28. Tool Call detect_fraud_patterns1.00
    29. Arguments:1.00
    30. 31.00
    31. Hybrid Search1.00
    32. "account_id": "6a5a92ab-0.97
    33. 7ela-4990-9ca0-0.99
    34. Semantic similarity (text embeddings) + structural similarity (FastRP graph embeddings)1.00
    35. fe08574b2ddc",0.99
    36. "neighbor_count": "1@"0.95
    37. Recommendation1.00
    38. ESCALATE FOR MANUAL REVIEW -0.97
    39. 41.00
    40. Recommendation1.00
    41. Do not approve automatically1.00
    42. Reasoning: While Jessica Norris has1.00
    43. strong financial indicators (substantial1.00
    44. account balance, low personal risk1.00
    45. score, clean transaction history), there1.00
    46. 31 nodes · 30 relationships · Jessica's complete context in a single graph traversal0.99
    47. are significant geopolitical and1.00

  • 7:43 #21 done

    shot 21·sharpness 2487.2

  • 7:49 #22 skipped

    shot 22·duplicate of #21

  • 8:20 #23 done

    shot 23·sharpness 3308.2

The page's on-screen-text budget of 600 lines is spent, so the last cards in this grid list fewer lines than they hold. Narrow the page with ?frames= to read them.

Transcript

159 cues· 2,638 words· 14,124 chars

  1. 0:14 So my name is Zach.
  2. 0:16 I work for Neo4j.
  3. 0:18 We're a graph intelligence company.
  4. 0:20 You can think of us like a knowledge layer graph database at the core.
  5. 0:24 We help connect and resolve information so AI systems can be a little bit more accurate and explainable.
  6. 0:31 I used to work in technical market.
  7. 0:32 I very recently am going back to my AI and machine learning roots.
  8. 0:37 transferred over to a research engineering role.
  9. 0:40 So I'm going to talk to you today about context graphs.
  10. 0:44 How many people in this room have heard about context graphs before?
  11. 0:49 Okay, so we've got about half.
  12. 0:50 And how many people have actually coded anything with a context graph or with a graph like Neo4j in general?
  13. 0:58 Okay, okay, about half, so that's good to know.
  14. 1:00 So I'm gonna talk a little bit about how we think about context graphs at Neo4j, define what they are, and then I want to go over some tools that are built by William Loyne, who's one of our product managers, so that you can get started with them very quickly with your own code and your agent framework.
  15. 1:17 So context graphs, foundation capital, I think back in December
  16. 1:26 And essentially, we saw a lot of noise around this idea of creating decision traces and reasoning to help agents make better decisions.
  17. 1:35 And so to kind of think actually about what a content graph is, you need to ask yourself, what do agents need to really be accurate, right?
  18. 1:43 And so for one thing, doing a lot of retrieval, you're going to need a knowledge base, obviously.
  19. 1:50 So a knowledge base for something like RAG or graph retrieval
  20. 1:53 basically helps an agent or a chatbot answer questions correctly.
  21. 2:00 What a context graph does, and the evolution of this, is really the information required to not only answer questions correctly, but make better decisions.
  22. 2:10 So if we took a concrete example, and I'll show you a demo of this, a financial analyst agent, and say that this agent has a question around improved interest rate increase and a request for a certain amount of money.
  23. 2:26 to allow that agent to receive information about the system.
  24. 2:30 So customer info, transactions, and policies.
  25. 2:34 The response from that is likely to look something like this, where it could maybe assign some sort of risk score and recommend some sort of review and talk about some key risk factors.
  26. 2:44 What is helpful, what a context graph enables an agent to do is actually give an answer.
  27. 2:52 Should you reject, accept, and why?
  28. 2:56 And it does this because in addition to getting that customer information in those transactions, it's also going to get you past decision traces and presidents, dynamic information about why decisions are made.
  29. 3:13 Systems of record, really about facts, entities, current state.
  30. 3:21 about presidents, causal chains, outcomes, and enabling the agent to act with subject matter expertise, and telling the agent really for a context graph, the entities about things that exist, there'll be a so decisions, transactions, approvals, things like that, and then content sees that
  31. 3:50 and different reasoning by AI that memory but by employees and past humans that have made some of it because you know with internet connection I never know you know how things are gonna go all the code for this is available I'll show links at the end basically and
  32. 4:18 Data that we generate, replicates, taking data in from a CRM and a support system in some other places.
  33. 4:25 It's using Claude in sort of the agent runtime, has some open AI embeddings.
  34. 4:31 There's Neo4j database assisting the data with some vectors.
  35. 4:36 And then we have our Next.js front end as well.
  36. 4:42 So let me show you this recording.
  37. 4:47 running here so you can kind of see what this looks like because again you ask it a question and it will call basically a series of tools and you can see it bringing the graph data back and then you see it getting these different decision traces and eventually it will get this reject decision right and if
  38. 5:27 Then it does this thing called find precedence, which is going to be in a second, but it's going to look at information to pull a bunch of
  39. 6:26 Internet would work.
  40. 6:30 Connect to each other in these.
  41. 6:31 Build off of each other.
  42. 6:53 is so like I was saying before, um, we query just context around Jessica's profile.
  43. 7:00 We pulled back decision traces and we did a special hybrid search, especially around those presidents, both on semantic similarity and structural similarity inside of the graph that actually looked at how the decision traces were made from previous decisions.
  44. 7:15 And it tried to match that.
  45. 7:17 And I'll talk a little bit about how that works with graph embeddings in the next slide to ultimately come up with the recommendation.
  46. 7:23 And so we add a, we have a vector index, so we're able to search like fraud rejection, for example, sort of on the semantics.
  47. 7:34 But then we have this called graph embedding.
  48. 7:36 So a lot of you in here are probably familiar with text embeddings on words.
  49. 7:42 A graph embedding is the same concept except those green nodes that I was showing you before everything was connected, we actually embedded those into a vector.
  50. 7:50 And so what that means is similar decision traces are now gonna be able to be looked up by vector similarity.

Open at this second