Skip to content

PHASE2_IMPLEMENTATION_STATUS.md

Latest commit

 

History

History
60 lines (45 loc) · 2.17 KB

PHASE2_IMPLEMENTATION_STATUS.md

File metadata and controls

60 lines (45 loc) · 2.17 KB

Phase 2: Performance Optimization - Implementation Status

Current Issues Found

Issue 1: Missing Layout Detection in Router Flow

Location: src/api/routers/document.py:1004-1010 Problem: OCR is called without layout_result parameter, but extract_text() requires it Impact: This may cause runtime errors or OCR may be using an empty dict Fix Required: Add layout detection before OCR, or make layout_result optional

Issue 2: Sequential Processing

Location: src/api/routers/document.py:998-1019 Problem: All stages run sequentially Impact: Slower processing time Optimization: Parallelize OCR + Layout Detection

Implementation Plan

Step 1: Fix Current Bug (CRITICAL - Must do first)

  • Add LayoutDetectionService to router flow
  • Ensure OCR receives proper layout_result
  • Test that current flow works correctly

Step 2: Parallel Processing (Low Risk)

  • Use asyncio.gather() to run OCR and Layout Detection in parallel
  • Both depend on preprocessing_result, but are independent
  • OCR can work with minimal/empty layout initially, then enhance later

Step 3: Caching (Medium Risk)

  • Create Redis cache service
  • Cache status queries (30s TTL)
  • Cache embeddings (24h TTL, content-hash based)
  • Cache API responses (1h TTL)

Step 4: Frontend Optimization (Higher Risk)

  • Implement SSE endpoint for real-time updates
  • Replace polling with SSE connection
  • Add exponential backoff fallback
  • Add optimistic UI updates

Files Created So Far

  1. ✅ docs/architecture/PHASE2_PERFORMANCE_PLAN.md - Overall plan
  2. ✅ src/api/services/document/parallel_executor.py - Parallel execution utilities
  3. ✅ docs/architecture/PHASE2_IMPLEMENTATION_STATUS.md - This file

Next Steps (Awaiting Approval)

  1. Fix Bug: Add layout detection to router flow
  2. Implement Parallel: Make OCR + Layout run in parallel
  3. Test: Verify no regressions
  4. Measure: Compare processing times

Risk Assessment

  • Step 1 (Fix Bug): Low risk, fixes existing issue
  • Step 2 (Parallel): Low risk, independent operations
  • Step 3 (Caching): Medium risk, need cache invalidation
  • Step 4 (Frontend): Higher risk, user-facing changes