[DEV] Integrate Langfuse Observability for Echo Agentic Workflows

[JasonMun7]

🔨 The Task

Incorporate Langfuse into the Echo backend (FastAPI) and frontend (Next.js) to enable full-stack tracing, cost monitoring, and performance evaluation.

🥅 Goal

We need to move beyond basic logging. Integrating Langfuse will allow us to:

1. Trace Complex Chains: Visualize the LangGraph execution flow in real-time.
2. Quality Scoring: Implement a system to score agent outputs (manual or LLM-as-a-judge) to improve synthesis accuracy.
3. Performance Bottlenecks: Identify which specific nodes in our agentic workflow are introducing the most latency.

📝 Implementation Details

• SDK Initialization: Install and configure langfuse and langfuse-python across the stack.
• LangGraph Tracing: Wrap the current LangGraph state machine with the Langfuse decorator/callback to capture trace data for every user request.
• Frontend Feedback Loop: Add a "thumbs up/down" or scoring component to the UI that pushes human-in-the-loop feedback directly to Langfuse.
• Metadata Tagging: Tag traces with user_id and session_id to correlate performance with specific Echo user contexts.

⚠️ Risk Level

Low
Langfuse operates asynchronously. If the Langfuse API is down, it should not block the primary Echo agent execution or response delivery.