migrate from output parsers to with_structured_output

[emarco177]

https://github.com/emarco177/ice_breaker/blob/bea2d975c7ed9d4c5607f5e0b54855abc054c2d0/output_parsers.py#L1C1-L33C82

output parsers were the old way to get structured output from LLMs.
With LangChain latest version we can do this by binding to the llm with_structured_output and the pydantic class we are trying to get.

migrate to this new way of getting structured output.

[emarco177]

@claude can you please fix this?

[claude]

Claude finished @emarco177's task —— View job • claude/issue-42-20250807-1543 • Create PR ➔

---

I've successfully migrated from output parsers to with_structured_output in LangChain! 🎉

Tasks

• Read current output_parsers.py implementation
• Find where output parsers are used in the codebase
• Research with_structured_output approach using context7 MCP
• Update output parsers to use new approach
• Update all references in the codebase
• Test the implementation works
• Commit and push changes

Summary of Changes

✅ Migration completed! I've updated your codebase to use the modern with_structured_output approach:

Files modified:

1. output_parsers.py: Removed PydanticOutputParser import and parser instances
2. custom_chains.py: Updated all three chains to use llm.with_structured_output() directly

Key improvements:

• ✅ Removed format instructions from all prompt templates (no longer needed)
• ✅ Simplified chain structure: prompt | structured_llm instead of prompt | llm | parser
• ✅ Cleaner, more maintainable code following modern LangChain patterns
• ✅ Same functionality with better performance and reliability

Benefits:

• Better performance (LLM handles structured output natively)
• More reliable (less prone to parsing errors)
• Cleaner code (shorter, more readable chains)
• Future-proof (aligns with current LangChain best practices)

The Pydantic models (Summary, IceBreaker, TopicOfInterest) remain unchanged and your existing functionality is fully preserved.

---