feat: define latency SLOs for AI story generation pipelinedocs: define latency SLOs

GroqTales/docs/PERFORMANCE.md

@@ -0,0 +1,104 @@
+\# Performance \& Latency Budget
+
+
+
+GroqTales defines Service Level Objectives (SLOs) for the AI story generation
+
+pipeline to ensure a fast and responsive user experience.
+
+
+
+This document outlines the expected latency targets across the full stack:
+
+UI → API → Groq → Database → UI Render.
+
+
+
+---
+
+
+
+\## 🎯 End-to-End Latency Target
+
+
+
+\*\*User Click → Story Fully Rendered\*\*
+
+
+
+\- \*\*Target (p95): < 4000ms\*\*
+
+
+
+This means 95% of users should see their complete story rendered within 4 seconds.
+
+
+
+---
+
+
+
+\## 🔎 Stage-Level Latency Targets
+
+
+
+| Stage | Description | Target (p95) |
+
+|--------|------------|--------------|
+
+| API Request Handling | Next.js API route processing time | < 100ms |
+
+| Groq API Call | Time taken for AI generation | < 2500ms |
+
+| First Token Streaming | Time until first token appears in UI | < 1000ms |
+
+| MongoDB Write | Time to persist story metadata | < 200ms |
+
+| UI Render Completion | Time to render full story in browser | < 300ms |
+
+
+
+---
+
+
+
+\## 📊 Why p95?
+
+
+
+We measure p95 (95th percentile) instead of average latency because
+
+average values can hide slow user experiences.
+
+
+
+p95 ensures that nearly all users experience acceptable performance,
+
+while allowing small network variations.
+
+
+
+---
+
+
+
+\## 📌 Purpose
+
+
+
+These SLOs will be used to:
+
+
+
+\- Monitor system performance
+
+\- Detect regressions
+
+\- Identify bottlenecks
+
+\- Power the internal Performance Dashboard
+
+
+
+
+