Skip to content

Tutorial 3#16

Merged
sattensil merged 49 commits intomainfrom
tutorial-3
Oct 10, 2025
Merged

Tutorial 3#16
sattensil merged 49 commits intomainfrom
tutorial-3

Conversation

@sattensil
Copy link
Copy Markdown
Collaborator

@sattensil sattensil commented Oct 2, 2025

tutorial- 3

To Review:

  • Metric tracking generally - Eng
  • Cost Implementation and custom metric - Eng/ DS
  • Experimentation - DS
  • Discussion of trends/ tracking and images to include - UX
  • Writing - DR

sattensil and others added 25 commits September 22, 2025 13:09
This extends the guided tutorial to part two which includes targeting and api implementation.
make supervisor intelligent
…metrics naturally

- Remove all manual recursion_limit calculations and enforcement
- Let LangGraph use natural recursion limits for proper ReAct agent reasoning
- LaunchDarkly's existing tool efficiency tracking preserved via ai_metrics/metrics_tracker.py
- Fixes support agent recursion limit errors while maintaining proper tool usage metrics
- Replace script-based setup with UI-guided LaunchDarkly experiment creation
- Add specific hypotheses with statistical thresholds (≥20% satisfaction, ≥25% tool efficiency)
- Include clear success/failure criteria and business impact calculations
- Focus on two core experiments: tool implementation ROI and model efficiency
- Revert traffic generator to realistic feedback simulation based on response quality
- Add statistical significance requirements and sample size guidance
- Replace frequentist statistics (ANOVA, p-values) with Bayesian posterior probabilities
- Add multi-variant experiment (4 variations: search_v1, search_v2+reranking, full stack, external only)
- Include triple-metric framework: satisfaction, cost constraint, latency threshold
- Update Claude model to claude-3-5-sonnet-20241022 specification
- Add comprehensive business impact calculations with net ROI formulas
- Replace statistical significance with >90% posterior probability thresholds
- Align analysis framework with LaunchDarkly's native Bayesian engine
Add comprehensive LangGraph documentation and educational comments
Message Memory Management/ Input Validation & Safety

Reliability improvements make the tutorial robust for long conversations
and handle edge cases gracefully while maintaining educational clarity.
Comment thread config_manager.py Outdated
Resolved conflicts by keeping tutorial-3 branch versions for all conflicting files.

Kept tutorial-3 versions:
- agents/ld_agent_helpers.py
- agents/security_agent.py
- agents/supervisor_agent.py
- api/main.py
- bootstrap/create_configs.py
- config_manager.py
- tools/traffic_generator.py
- ui/chat_interface.py

Kept deletions from tutorial-3:
- ai_metrics/metrics_tracker.py
- ai_metrics/tool_performance.py
- data/fake_users.json
- data/sample_queries.json

Excluded debug/test files not needed for tutorial-3.
@sattensil sattensil merged commit cf13775 into main Oct 10, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants