Match Analysis AI Agent Fails with Parser Error

[aliciusschroeder]

Description

The current AI match analysis endpoint (/match/ai/:matchId) consistently fails with a parsing error when attempting to analyze League of Legends match data. The error suggests that the LangChain agent is unable to properly interpret and process the match data structure.

Current Behavior

When making a request to /match/ai/:matchId, the endpoint returns a 500 error with the following stack trace:

Error: Could not parse LLM output: I don't know
    at ZeroShotAgentOutputParser.parse (node_modules/langchain/dist/agents/mrkl/outputParser.cjs:29:19)
    at ZeroShotAgent._plan (node_modules/langchain/dist/agents/agent.cjs:205:34)
    at processTicksAndRejections (node:internal/process/task_queues:95:5)
    at AgentExecutor._call (node_modules/langchain/dist/agents/executor.cjs:85:28)
    at AgentExecutor.call (node_modules/langchain/dist/chains/base.cjs:68:28)

Expected Behavior

The endpoint should analyze the match data and return meaningful insights about:

• Match duration and outcome
• Team performance metrics
• Player performance statistics
• Objective control (dragons, towers, etc.)

Reproduction Steps

1. Start the server with valid environment variables
2. Make a GET request to /match/ai/{any-valid-match-id}
3. Observe the parsing error in the response

Context

• The raw match data from Riot's API is being successfully fetched (verified through /match/:matchId endpoint)
• The error occurs during the LangChain agent's attempt to analyze this data
• The match endpoint in the application is functioning correctly
• The issue appears to be specific to the AI analysis implementation

Environment

• Node.js version: 20.11.1
• TypeScript version: 5.6.3
• LangChain version: 0.0.71
• Operating System: Windows

Additional Notes

• The error is consistent across different match IDs
• The current prompt may be too generic for the complex structure of League match data
• The JSON structure from Riot's API contains deeply nested match information that may need more specific handling

Related Components

• src/services/analyze-league-match.ts
• src/routes/match/ai.ts
• LangChain's JsonToolkit and JsonSpec utilities

Possible Areas of Investigation

1. The structure and specificity of the AI agent's prompt
2. How the JsonToolkit handles complex, nested game data
3. The interaction between the agent and League's match data format
4. Alternative approaches to match data analysis that might be more reliable

Impact

This issue blocks the core functionality of AI-powered match analysis, which is a key feature of the application.