Stop AI Agent Hallucinations: 5 Techniques + Production Patterns - Elizabeth Fuentes, AWS

What was stored

cues

538

whisperx 538

chunks

101

from 538 cues

keyframes

384

kept of 443 captured

frames with text

369

16,992 lines read

chapters

0

from the source metadata

keyframe bytes

70.4 MB

word timings on 538 cues

Provenance

Frames, and what the machine read

• 

  0:12 #0 empty

  shot 0·sharpness 117.2

• 

  0:33 #1 done7 line(s)

  shot 1·sharpness 1282.2

  1. Stop Al Agent Hallucinations1.00
  2. 5 Techniques + Production Patterns1.00
  3. Five techniques beyond the prompt — each one a code change, not a prompt1.00
  4. change1.00
  5. Demos built with Strands Agents · Deployable on Amazon Bedrock AgentCore1.00
  6. aws0.87
  7. © 2026, Amazon Web Services, Inc. or its affiliates. All rights reserved.0.99
• 

  1:18 #2 done27 line(s)

  shot 2·sharpness 1828.9

  1. Five Techniques Beyond the Prompt0.99
  2. Each fixes a different failure. None of them is a prompt problem.0.99
  3. 21.00
  4. 31.00
  5. 41.00
  6. 51.00
  7. Semantic1.00
  8. Multi-Agent1.00
  9. Neurosymbolic1.00
  10. Graph-RAG1.00
  11. Runtime Steering1.00
  12. Tool Selection1.00
  13. Validation1.00
  14. Guardrails1.00
  15. Filter tools into1.00
  16. Compute precise1.00
  17. A second agent checks1.00
  18. Rules in code, not in1.00
  19. Self-correct instead of0.98
  20. context on every call1.00
  21. answers, don't guess1.00
  22. every response1.00
  23. the prompt1.00
  24. blocking1.00
  25. ...then ship all five to production with Amazon Bedrock AgentCore0.99
  26. aws0.99
  27. © 2026, Amazon Web Services, Inc. or its affiliates. All rights reserved.0.99
• 

  1:40 #3 skipped

  shot 3·duplicate of #2

• 

  2:04 #4 skipped

  shot 4·duplicate of #2

• 

  2:36 #5 skipped

  shot 5·duplicate of #2

• 

  3:02 #6 done8 line(s)

  shot 6·sharpness 1530.3

  1. aws0.99
  2. Elizabeth Fuentes Leone1.00
  3. Developer Advocate @ AWS0.99
  4. elifuentes.tech1.00
  5. Resources1.00
  6. bit.ly/4oN2LVu1.00
  7. aws1.00
  8. ©2026, Amazon Web Services, Inc. or its affiliates. All rights reserved.0.99
• 

  3:50 #7 empty

  shot 7·sharpness 91.2

• 

  4:34 #8 empty

  shot 8·sharpness 93.6

• 

  4:58 #9 done22 line(s)

  shot 9·sharpness 2291.3

  1. 1 Semantic Tool Selection0.99
  2. First: what a tool actually looks like to the model1.00
  3. You write a @tool1.00
  4. Strands generates a schema1.00
  5. Into context on every call0.99
  6. @tool1.00
  7. def search_hotels(query: str) ->0.98
  8. "name": ...0.91
  9. ~70–100 tokens per schema0.99
  10. str:1.00
  11. "description": ...0.95
  12. """Search hotels by1.00
  13. city or location."""0.98
  14. "parameters": ..0.93
  15. × dozens of tools1.00
  16. = thousands of tokens1.00
  17. every call, before your message1.00
  18. name·docstring·typed params1.00
  19. the model reads this, not your code1.00
  20. If your agent has memory, that grows too — every turn adds context that ships with every message.0.99
  21. aws1.00
  22. © 2026, Amazon Web Services, Inc. or its affiliates. All rights reserved.0.99
• 

  5:11 #10 skipped

  shot 10·duplicate of #9

• 

  5:39 #11 skipped

  shot 11·duplicate of #9

• 

  6:07 #12 done64 line(s)

  shot 12·sharpness 3976.2

  1. EXPLORER1.00
  2. 0000.76
  3. test_graphrag.ipynb1.00
  4. token_efficiency_analysis.ipynb1.00
  5. test_semantic_tools_hallucinations.ipynb1.00
  6. 0000.77
  7. WHY-AGENTS-FAIL-SAMPLE-...0.99
  8. -tools-demo>token_efficiency_analysis.ipynb >M↓ Semantic Tool Selection: Token Efficiency & Accuracy Analysis > M↓ Build Semantic Index>build_index(ALL_TOOLS)0.98
  9. Q0.99
  10. > .claude0.89
  11. Generate + Code + MarkdownRun All Restart Clear All OutputsJupyter Variables Outline0.97
  12. 0000.97
  13. Python 3.12.131.00
  14. > 01-faq-graphrag-demo0.97
  15. 02-semantic-tools-demo1.00
  16. < Back to Main README | Demo README0.97
  17. __pycache_0.90
  18. > images0.92
  19. enhanced_tools.py1.00
  20. Semantic Tool Selection: Token Efficiency & Accuracy Analysis1.00
  21. registry.py1.00
  22. ① README.md0.96
  23. 业0.92
  24. Based on: Internal Representations as Indicators of Hallucinations in Agent Tool Selection0.99
  25. requirements.txt0.98
  26. test_semantic_tools_hallucinations.ipynb1.00
  27. What This Demo Measures1.00
  28. token_comparison_app.py1.00
  29. 00.89
  30. token_efficiency_analysis.ipynb1.00
  31. > 03-multiagent-demo0.97
  32. This notebook compares Traditional (all 31 tools) vs Semantic (top-3 filtered tools) approaches across two critical metrics:1.00
  33. 30.55
  34. > 04-neurosymbolic-demo0.98
  35. 1. Token Consumption: How many tokens are used per query?0.99
  36. > 05-agent-control-demo0.97
  37. 2. Tool Selection Accuracy: Does the agent pick the correct tool?1.00
  38. aws1.00
  39. 07-context-graph-integration1.00
  40. 06-agentcore-production-demo0.99
  41. The Dual Problem0.99
  42. > artifacts0.97
  43. > blog-es0.98
  44. • Token Waste: Sending 31 tool descriptions = ~4,500 tokens per query0.98
  45. > images0.93
  46. Hallucination Risk: More tools = more confusion = wrong tool selection0.99
  47. langgraph-demos1.00
  48. .gitignore1.00
  49. The Solution1.00
  50. CLAUDE.md1.00
  51. CODE_OF_CONDUCT.md1.00
  52. User Query → FAISS Search → Top 3 Tools → Agent → Correct Selection + Fewer Tokens0.99
  53. LICENSE0.97
  54. ③ README.md0.93
  55. test_nova_embeddings.py1.00
  56. !pip install -q -r requirements.txt0.95
  57. [13]1.00
  58. 2.3s1.00
  59. 0000.95
  60. [notice] A new release of pip is available: 25.0.1 -> 26.1.20.98
  61. [notice] To update, run: pip install --upgrade pip0.99
  62. Configure API Key1.00
  63. OUTLINE0.96
  64. > APPLICATION BUILDER0.96
• 

  6:21 #13 done63 line(s)

  shot 13·sharpness 3350.5

  1. EXPLORER1.00
  2. 0000.74
  3. test_graphrag.ipynb1.00
  4. token_efficiency_analysis.ipynb0.99
  5. test_semantic_tools_hallucinations.ipynb1.00
  6. WHY-AGENTS-FAIL-SAMPLE-...0.99
  7. -tools-demo>token_efficiency_analysis.ipynb> M↓ Semantic Tool Selection: Token Efficiency & Accuracy Analysis > M↓ What This Demo Measures > M↓ The Solution >0.97
  8. Q0.99
  9. > .claude0.90
  10. Generate + Code + MarkdownRun All Restart Clear All OutputsJupyter Variables Outline0.97
  11. 0001.00
  12. Python 3.12.130.99
  13. > 01-faq-graphrag-demo0.97
  14. 02-semantic-tools-demo1.00
  15. This notebook compares Traditional (all 31 tools) vs Semantic (top-3 filtered tools) approaches across two critical metrics:1.00
  16. >__pycache__0.85
  17. 1. Token Consumption: How many tokens are used per query?0.99
  18. 好0.65
  19. > images0.94
  20. 2. Tool Selection Accuracy: Does the agent pick the correct tool?1.00
  21. enhanced_tools.py1.00
  22. ①README.md0.98
  23. The Dual Problem1.00
  24. registry.py1.00
  25. test_semantic_tools_hsliucinations.ipynb0.98
  26. requirements.txt0.98
  27. • Token Waste: Sending 31 tool descriptions = ~4,500 tokens per query0.98
  28. • Hallucination Risk: More tools = more confusion = wrong tool selection0.99
  29. token_comparison_app.p ~/Documents/repositories/eli-cosas/why-agents-fail-sample-for-amazon-agentcore/02-0.99
  30. 00.86
  31. token_efficiency_analysis semantic-tools-demo/test_semantic_tools_hallucinations.ipynb1.00
  32. > 03-multiagent-demo0.97
  33. >04-neurosymbolic-demo0.99
  34. User Query → FAISS Search → Top 3 Tools → Agent → Correct Selection + Fewer Tokens0.99
  35. > 05-agent-control-demo0.97
  36. θ0.62
  37. aws1.00
  38. >07-context-graph-integration0.99
  39. 06-agentcore-production-demo1.00
  40. !pipinstall -q -r requirements.txt0.96
  41. > artifacts0.96
  42. Python1.00
  43. > blog-es0.96
  44. > images0.94
  45. Configure API Key1.00
  46. langgraph-demos1.00
  47. .gitignore1.00
  48. CLAUDE.md1.00
  49. LICENSE0.99
  50. CODE_OF_CONDUCT.md1.00
  51. from dotenv import load_dotenv0.99
  52. import os1.00
  53. ①README.md0.94
  54. load_dotenv()1.00
  55. test_nova_embeddings.py1.00
  56. U1.00
  57. # Uncomment and set your key if not using a .env file:0.99
  58. # os.environ['OPENAI_API_KEY'] = 'your-key-here'0.97
  59. assert os.getenv('OPENAI_API_KEY'), 'Set OPENAI_API_KEY in .env file or uncomment line above'1.00
  60. []0.99
  61. Setup1.00
  62. > OUTLINE0.91
  63. APPLICATION BUILDER0.99
• 

  6:47 #14 done62 line(s)

  shot 14·sharpness 1722.8

  1. EXPLORER1.00
  2. 0000.78
  3. test_graphrag.ipynb1.00
  4. token_efficiency_analysis.ipynb1.00
  5. requirements.txt ×0.96
  6. test_semantic_tools_hallucinations.ipynb1.00
  7. ①0.99
  8. 0000.63
  9. WHY-AGENTS-FAIL-SAMPLE-FOR-AMAZON-AG...1.00
  10. 02-semantic-tools-demo requirements.txt0.99
  11. Q0.99
  12. .claude0.97
  13. > 01-faq-graphrag-demo0.96
  14. Search files by name (append : to go to line or @ to go to symbol)0.99
  15. 990.70
  16. 02-semantic-tools-demo1.00
  17. Go to File1.00
  18. 960.99
  19. __pycache__0.91
  20. Show and Run Commands >0.99
  21. > images0.93
  22. Search for Text %1.00
  23. enhanced_tools.py1.00
  24. ① README.md0.95
  25. Open Quick Chat0.99
  26. registry.py1.00
  27. Go to Symbol in Editor @1.00
  28. test_semantic_tools_hallucinations.ipynb1.00
  29. requirements.txt0.98
  30. Run Task task0.96
  31. Start Debugging debug1.00
  32. token_comparison_app.py1.00
  33. 00.57
  34. token_efficiency_analysis.ipynb1.00
  35. More ?0.92
  36. I0.87
  37. > 03-multiagent-demo0.96
  38. requirements.txt 02-semantic-tools-demo0.98
  39. recently opened1.00
  40. 30.53
  41. > 04-neurosymbolic-demo0.97
  42. token_efficiency_analysis.ipynb 02-semantic-tools-demo0.99
  43. > 05-agent-control-demo0.96
  44. aws1.00
  45. > 06-agentcore-production-demo0.98
  46. test_semantic_tools_hallucinations.ipynb 02-semantic-tools-demo1.00
  47. >07-context-graph-integration0.98
  48. test_graphrag.ipynb 01-faq-graphrag-demo0.99
  49. > artifacts0.95
  50. .env 01-faq-graphrag-demo1.00
  51. > blog-es0.98
  52. > images0.91
  53. langgraph-demos1.00
  54. .gitignore1.00
  55. CLAUDE.md1.00
  56. CODE_OF_CONDUCT.md1.00
  57. LICENSE0.98
  58. README.md1.00
  59. test_nova_embeddings.py1.00
  60. U0.99
  61. OUTLINE0.95
  62. APPLICATION BUILDER0.99
• 

  7:01 #15 done46 line(s)

  shot 15·sharpness 1353.6

  1. EXPLORER1.00
  2. 0000.87
  3. test_graphrag.ipynb1.00
  4. token_efficiency_analysis.ipynb1.00
  5. requirements.txt X0.96
  6. test_semantic_tools_hallucinations.ipynb1.00
  7. 0000.68
  8. WHY-AGENTS-FAIL-SAMPLE-FOR-AMAZON-AG...1.00
  9. 02-semantic-tools-demo> requirements.txt0.98
  10. .claude0.98
  11. strands-agents[openai]>=1.27.01.00
  12. > 01-faq-graphrag-demo0.97
  13. sentence-transformers>=3.4.01.00
  14. 990.72
  15. 02-semantic-tools-demo1.00
  16. faiss-cpu>=1.9.01.00
  17. neo4j>=5.28.01.00
  18. __pycache_0.90
  19. python-dotenv>=1.0.11.00
  20. > images0.94
  21. enhanced_tools.py1.00
  22. ① README.md0.94
  23. registry.py1.00
  24. test_semantic_tools_hallucinations.ipynb1.00
  25. requirements.txt0.98
  26. token_comparison_app.py1.00
  27. token_efficiency_analysis.ipynb1.00
  28. I0.92
  29. >03-multiagent-demo0.98
  30. > 04-neurosymbolic-demo0.98
  31. > 05-agent-control-demo0.97
  32. aws1.00
  33. > 06-agentcore-production-demo0.98
  34. > 07-context-graph-integration0.98
  35. artifacts0.98
  36. > blog-es0.92
  37. > images0.92
  38. langgraph-demos1.00
  39. .gitignore1.00
  40. CLAUDE.md1.00
  41. CODE_OF_CONDUCT.md1.00
  42. LICENSE0.97
  43. README.md1.00
  44. test_nova_embeddings.py1.00
  45. OUTLINE0.95
  46. APPLICATION BUILDER0.97
• 

  7:47 #16 skipped

  shot 16·duplicate of #15

• 

  7:59 #17 done79 line(s)

  shot 17·sharpness 3296.5

  1. EXPLORER1.00
  2. 0000.82
  3. test_graphrag.ipynb1.00
  4. token_efficiency_analysis.ipynb1.00
  5. requirements.txt0.97
  6. test_semantic_tools_hallucinations.ipynb1.00
  7. 00。0.64
  8. WHY-AGENTS-FAIL-SAMPLE-FOR-AMAZON-AG...0.99
  9. 02-semantic-tools-demo >test_semantic_tools_hallucinations.ipynb > M↓ Semantic Tool Discovery: Reducing Tool Selection Hallucinations0.98
  10. .claude0.97
  11. Generate + Code + MarkdownRun All Restart Clear All Outputs0.97
  12. Go ToJupyter Variables Outline0.99
  13. 0000.97
  14. Python 3.10.201.00
  15. > 01-faq-graphrag-demo0.97
  16. 02-semantic-tools-demo1.00
  17. e0.56
  18. import io1.00
  19. import sys0.99
  20. __pycache__0.90
  21. import re0.97
  22. > images0.90
  23. e0.64
  24. enhanced_tools.py1.00
  25. from strands import Agent1.00
  26. registry.py1.00
  27. ①README.md0.99
  28. from registry import build_index, search_tools, get_scores1.00
  29. from enhanced_tools import ALL_ToOLS0.97
  30. test_semantic_tools_hallucinations.ipynb1.00
  31. requirements.txt0.99
  32. [3]1.00
  33. print(f" Loaded {len(ALL_ToOLS)} tools")0.97
  34. 1.9s1.00
  35. Python1.00
  36. token_comparison_app.py1.00
  37. 00.88
  38. token_efficiency_analysis.ipynb0.99
  39. 0000.97
  40. ModuleNotFoundError1.00
  41. Traceback (most recent call last)1.00
  42. > 03-multiagent-demo0.96
  43. Cell In[3], line 70.97
  44. > 04-neurosymbolic-demo0.97
  45. 5 from strands import Agent0.97
  46. > 05-agent-control-demo0.97
  47. 7 from registry import build_index, search_tools, get_scores0.99
  48. 6 from enhanced_tools import ALL_T0OLS0.97
  49. aws0.89
  50. > 06-agentcore-production-demo0.98
  51. 9 print(f" Loaded {len(ALL_T00LS)} tools")0.96
  52. > 07-context-graph-integration0.98
  53. > artifacts0.94
  54. File ~/Documents/repositories/eli-cosas/why-agents-fail-sample-for-amazon-agentcore/02-semantic-tools-demo/registry-py:41.00
  55. > blog-es0.97
  56. 10.84
  57. > images0.93
  58. 2 Tool Registry - FAISS semantic search over tools0.99
  59. 3 "……0.63
  60. langgraph-demos0.97
  61. --> 4 import faiss0.97
  62. .gitignore1.00
  63. 5 from sentence_transformers import SentenceTransformer0.99
  64. CLAUDE.md1.00
  65. 6 from typing import List, Callable0.97
  66. CODE_OF_CONDUCT.md1.00
  67. LICENSE0.97
  68. ModuleNotFoundError: No module named 'faiss'1.00
  69. ③ README.md0.91
  70. test_nova_embeddings.py1.00
  71. U0.99
  72. Build Semantic Index1.00
  73. For each tool, we concatenatename: docstringand encode it withall-MiniLM-L6-v2 (384-dim vectors). FAISs0.99
  74. L2 nearest-neighbor search. The score shown is 1/(1+distance)- closer to 1.0 means better match.0.98
  75. build_index(ALL_T00LS)0.98
  76. OUTLINE1.00
  77. # Test semantic search0.98
  78. > APPLICATION BUILDER0.95
  79. []0.97
• 

  7:59 #18 done47 line(s)

  shot 18·sharpness 2334.9

  1. test_graphrag.ipynb1.00
  2. token_efficiency_analysis.ipynb1.00
  3. requirements.txt1.00
  4. test_semantic_tools_hallucinations.ipynb1.00
  5. ①0.98
  6. 00。0.64
  7. 02-semantic-tools-demo >test_semantic_tools_hallucinations.ipynb > M↓ Semantic Tool Discovery: Reducing Tool Selection Hallucinations0.98
  8. Q0.99
  9. Generate + Code + Markdown Run All Restart Clear All Outputs Go To Jupyter Variables ≡Outline0.96
  10. Python 3.10.201.00
  11. D0.98
  12. import sys1.00
  13. I0.71
  14. import io1.00
  15. import re1.00
  16. 好0.59
  17. from strands import Agent0.99
  18. from enhanced_tools import ALL_TOOLS0.97
  19. from registry import build_index, search_tools, get_scores0.99
  20. print(f" Loaded {len(ALL_ToOLS)} tools")0.97
  21. [3]1.00
  22. 1.9s1.00
  23. Python1.00
  24. 00.94
  25. 0000.82
  26. ModuleNotFoundError1.00
  27. Traceback (most recent call last)1.00
  28. Cell In[3], line 70.98
  29. 5 from strands import Agent1.00
  30. 6 from enhanced_tools import ALL_ToOLS0.97
  31. 7 from registry import build_index, search_tools, get_scores1.00
  32. aws1.00
  33. 9 print(f" Loaded {len(ALL_To0LS)} tools")0.97
  34. File ~/Documents/repositories/eli-cosas/why-agents-fail-sample-for-amazon-agentcore/02-semantic-tools-demo/registry.py:40.99
  35. 1 "……0.70
  36. 2 Tool Registry - FAISS semantic search over tools0.99
  37. 3 "……0.66
  38. --> 4 import faiss0.94
  39. 5 from sentence_transformers import SentenceTransformer1.00
  40. 6 from typing import List, Callable0.98
  41. ModuleNotFoundError: No module named 'faiss'1.00
  42. Build Semantic Index1.00
  43. For each tool, we concatenate name: docstring and encode it with all-MiniLM-L6-v2 (384-dim vectors). FAISS stores these vectors and performs L2 nearest0.99
  44. is1/(1+distance) - closer to 1.0 means better match.0.98
  45. build_index(ALL_T00LS)0.97
  46. # Test semantic search0.98
  47. []0.98
• 

  8:02 #19 done39 line(s)

  shot 19·sharpness 4312.3

  1. why-agents-fail-sample-for-amazon-agentcore1.00
  2. U80.83
  3. oken_efficiency_analysis.ipynb1.00
  4. Erequirements.txt0.97
  5. test_semantic_tools_hallucinations.ipynb1.00
  6. o0o0.59
  7. 02-semantic-tools-demo >0.98
  8. test_semantic_tools_hallucinations.ipynb > M↓ Semantic Tool Discovery: Reducing Tool Select0.98
  9. Generate1.00
  10. + Code0.94
  11. + Markdown0.96
  12. Run All0.95
  13. GRestart0.91
  14. Clear All Outputs1.00
  15. 0000.98
  16. Python 3.10.201.00
  17. import sys0.99
  18. import io1.00
  19. 961.00
  20. import re1.00
  21. 深0.51
  22. from strands import Agent1.00
  23. from enhanced_tools import ALL_To0LS0.97
  24. 田0.61
  25. from registry import build_index, search_tools, get_scores0.99
  26. print(f" Loaded {len(ALL_TOOLS)} tools")0.97
  27. [3]1.00
  28. 1.9s1.00
  29. Python1.00
  30. 0000.97
  31. 0oo0.56
  32. ModuleNotFoundError1.00
  33. Traceback (most recent call last)1.00
  34. Cell In[3], line 70.95
  35. 5 from strands import Agent0.99
  36. 6 from enhanced_tools import ALL_ToOLS0.97
  37. 7 from registry import build_index, search_tools, get_scores0.99
  38. 9 print(f" Loaded {len(ALL_TOOLS)} tools")0.97
  39. File ~/Documents/repositories/eli-cosas/why-agents-fail-sample-for-amazon-agentcc1.00
• 

  8:06 #20 done43 line(s)

  shot 20·sharpness 2997.3

  1. test_graphrag.ipynb1.00
  2. token_efficiency_analysis.ipynb1.00
  3. requirements.txt1.00
  4. test_semantic_tools_hallucinations.ipynb0.99
  5. 0000.95
  6. 2-semantic-tools-demo >test_semantic_tools_hallucinations.ipynb> M↓ Semantic Tool Discovery: Reducing Tool Selection Hallucinations0.99
  7. Generate + Code + Markdown0.97
  8. Run All Restart Clear All Outputs0.97
  9. Go ToJupyter Variables Outline0.99
  10. 0000.86
  11. Python 3.10.201.00
  12. import sys1.00
  13. 290.70
  14. import io1.00
  15. 961.00
  16. import re0.94
  17. from strands import Agent1.00
  18. from enhanced_tools import ALL_ToOLS0.98
  19. 田0.66
  20. from registry import build_index, search_tools, get_scores0.98
  21. print(f" Loaded {len(ALL_ToOLS)} tools")0.96
  22. [3]1.00
  23. 区0.63
  24. 1.9s0.91
  25. Python1.00
  26. 0001.00
  27. ModuleNotFoundError1.00
  28. Traceback (most recent call last)0.99
  29. Cell In[3], line 70.97
  30. 5 from strands import Agent1.00
  31. 6 from enhanced_tools import ALL_ToOLS0.97
  32. 7 from registry import build_index, search_tools, get_scores0.99
  33. aws1.00
  34. 9 print(f" Loaded {len(ALL_TO0LS)} tools")0.96
  35. File ~/Documents/repositories/eli-cosas/why-agents-fail-sample-for-amazon-agentcore/02-semantic-tools-demo/registry.py:41.00
  36. 1"…"0.85
  37. 2 Tool Registry - FAISS semantic search over tools0.98
  38. 3 "……0.78
  39. 4 import faiss1.00
  40. 5 from sentence_transformers import SentenceTransformer0.99
  41. 6 from typing import List, Callable1.00
  42. ModuleNotFoundError: No module named 'faiss'0.98
  43. Build Semantic Index0.98
• 

  8:07 #21 done73 line(s)

  shot 21·sharpness 4325.6

  1. EXPLORER1.00
  2. 0000.99
  3. token_efficiency_analysis.ipynb1.00
  4. requirements.txt1.00
  5. test_semantic_tools_hallucinations.ipynb1.00
  6. 0000.92
  7. WHY-AGENTS-FAIL-SAMPLE-...0.99
  8. 02-semantic-tools-demo >test_semantic_tools_hallucinations.ipynb >M↓ Semantic Tool Discovery: Reducing Tool Selection Halluci0.98
  9. .claude1.00
  10. Generate + Code + MarkdownRun All Restart Clear All Outputs0.97
  11. Go To1.00
  12. 0000.93
  13. Python 3.10.201.00
  14. >01-faq-graphrag-demo0.98
  15. import sys1.00
  16. 290.83
  17. 02-semantic-tools-demo1.00
  18. import io1.00
  19. 961.00
  20. __pycache__0.88
  21. import re1.00
  22. images1.00
  23. enhanced_tools.py1.00
  24. from strands import Agent1.00
  25. 60.51
  26. README.md1.00
  27. from registry import build_index, search_tools, get_scores1.00
  28. from enhanced_tools import ALL_ToOLS0.97
  29. registry.py0.98
  30. requirements.txt0.97
  31. print(f" Loaded {len(ALL_ToOLS)} tools")0.96
  32. test_semantic_tools_hallucinations.ipynb1.00
  33. [3]1.00
  34. X0.60
  35. 1.9s1.00
  36. Python1.00
  37. token_comparison_app.py1.00
  38. token_efficiency_analysis.ipynb1.00
  39. 0000.99
  40. ModuleNotFoundError1.00
  41. Traceback (most recent call last)1.00
  42. >03-multiagent-demo0.98
  43. Cell In[3], line 70.97
  44. > 04-neurosymbolic-demo0.97
  45. 5 from strands import Agent0.99
  46. > 05-agent-control-demo0.97
  47. 6 from enhanced_tools import ALL_T0OLS0.96
  48. 7 from registry import build_index, search_tools, get_scores0.99
  49. aws1.00
  50. > 06-agentcore-production-demo0.97
  51. 9 print(f" Loaded {len(ALL_TO0LS)} tools")0.95
  52. > 07-context-graph-integration0.98
  53. > artifacts0.97
  54. File ~/Documents/repositories/eli-cosas/why-agents-fail-sample-for-amazon-agentcore/02-semantic-tools-demo/t1.00
  55. > blog-es0.93
  56. 1 "…"0.73
  57. > images0.96
  58. 2 Tool Registry - FAISS semantic search over tools0.98
  59. 3 "…"0.74
  60. > langgraph-demos0.95
  61. 4 import faiss0.97
  62. .gitignore1.00
  63. 5 from sentence_transformers import SentenceTransformer0.99
  64. CLAUDE.md1.00
  65. 6 from typing import List, Callable0.99
  66. CODE_OF_CONDUCT.md1.00
  67. LICENSE1.00
  68. ModuleNotFoundError: No module named 'faiss'0.98
  69. ①README.md0.95
  70. > OUTLINE0.89
  71. test_nova_embeddings.py1.00
  72. Build Semantic Index0.99
  73. APPLICATION BUILDER0.97
• 

  8:09 #22 done20 line(s)

  shot 22·sharpness 4541.3

  1. EXPLORER1.00
  2. 0000.99
  3. token_efficiency_analysis.ipynb1.00
  4. requirements.txt1.00
  5. test_semantic_tools_hallucinations.ipynb0.99
  6. 0000.92
  7. WHY-AGENTS-FAIL-SAMPLE-FOR-AMAZON-AG...1.00
  8. 02-semantic-tools-demo>token_efficiency_analysis.ipynb >M↓ Semantic Tool Selection: Token Efficiency & Accuracy Analysis >MJ0.98
  9. .claude1.00
  10. Generate +Code +MarkdownRun All Restart Clear All Outputs0.97
  11. 0000.96
  12. Python 3.12.131.00
  13. 01-faq-graphrag-demo1.00
  14. This notebook compares Traditional (all 31 tools) vs Semantic (top-3 filtered tools) approaches across two critical1.00
  15. 290.82
  16. 02-semantic-tools-demo1.00
  17. metrics:1.00
  18. 961.00
  19. __pycache__0.86
  20. 1. Token Consumption: How many tokens are used per query?1.00
• 

  8:12 #23 done

  shot 23·sharpness 3270.2

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

538 cues· 6,909 words· 36,062 chars