Skip to content

errors.md

Latest commit

 

History

History
112 lines (87 loc) · 4.27 KB

errors.md

File metadata and controls

112 lines (87 loc) · 4.27 KB
Raw

Errors and Lessons Learned - LangGraph Joke Bot Project

This document records the errors encountered during development of the Joke Bot project using LangGraph, along with the lessons learned from each issue.

1. Module Import Errors

Error:

ModuleNotFoundError: No module named 'frontend'

Cause: Attempted to import a non-existent or misplaced module.
Lesson Learned: Always verify the project structure and ensure imports match the actual folder/file organization.

2. FileNotFoundError for prompt_config.yaml

Error:

FileNotFoundError: [Errno 2] No such file or directory: '.../config/prompt_config.yaml'

Cause: The program looked for prompt_config.yaml in the wrong directory.
Lesson Learned: Keep configuration files in a clearly defined path and verify relative vs. absolute paths. Use os.path.join to avoid mistakes.

3. API Connection Issues

Error:

requests.exceptions.ConnectionError: HTTPSConnectionPool(host='research-assistant-oe9n.onrender.com', ...)

Cause: The host could not be resolved or internet connection was unstable.
Lesson Learned: Confirm that APIs are deployed and reachable. Use ping or curl to verify connectivity before debugging code.

4. Misconfigured Prompt (Wrong Field Injection)

Issue: Jokes did not use the selected language or category.
Cause: Prompt construction used placeholders incorrectly, e.g.:

prompt += f"\n\nThe category is: {state.language}"
prompt += f"\n\nThe category is: {state.category}"

This caused confusion in output.
Lesson Learned: Be precise in prompt building, and ensure variables like language and category are injected in the right context.

5. Writer–Critic Loop Not Triggering Properly

Issue: The writer-critic loop didn’t always repeat.
Cause: Router logic was incomplete:

if state.approved or state.retry_count >= 5:
    return "show_final_joke"
return "writer"

Lesson Learned: Carefully define routing edges to enforce desired behavior. Test each condition with multiple inputs.

6. Misunderstanding Between Language vs Category

Issue: Language selection worked, but the joke generator treated language as category.
Lesson Learned: Clearly separate and name state fields (e.g., state.language, state.category) to avoid overlap.

7. Pinecone Integration Slowness

Issue: Program took very long to start or save jokes.
Cause: Pinecone client and index were being reinitialized multiple times instead of once.
Lesson Learned: Initialize Pinecone client and index once globally, and reuse them across nodes to reduce overhead.

8. AttributeError: Pinecone client

Error:

AttributeError: 'Pinecone' object has no attribute 'index'. Did you mean: 'Index'?

Cause: Used the wrong method (pc.index) instead of pc.Index.
Lesson Learned: Double-check Pinecone SDK methods—current versions require pc.Index("name").

9. KeyError: 'browse_jokes'

Error:

KeyError: 'browse_jokes'
During task with name 'show_menu'

Cause: A menu option (b) mapped to browse_jokes, but the node was not registered in the graph’s routing.
Lesson Learned: Always ensure new menu actions are added both as nodes and edges in the graph definition.

10. Graph Visualization Not Showing

Issue: Wanted to see the graph, but nothing was rendered.
Cause: Did not generate or open the graph.mmd file.
Lesson Learned: Use graph.get_graph().draw_mermaid() to export to .mmd, and view with mermaid.live or VS Code’s Mermaid preview extension.

General Lessons

  • Always keep configs in sync with your codebase.
  • Add logging (print or proper logger) to inspect the state at each node.
  • Test small components (like router logic) independently before wiring the full graph.
  • Be explicit in prompt engineering: the LLM outputs only what you ask it to.
  • Optimize for performance: reuse Pinecone client and embedding models instead of creating them repeatedly.
  • When extending functionality (like adding browse_jokes), update both the state machine and the graph connections.

With each error fixed, the system became more robust and predictable!