Enhance Resume Tailoring System: Connect n8n to FAISS + Schema Enforcement

[BPMSoftwareSolutions]

Overview

The resume tailoring system has a mature, production-grade RAG infrastructure (FAISS + sentence-transformers) that is currently disconnected from the n8n workflows. The n8n workflows use simple keyword matching instead of semantic search, resulting in lower-quality resume tailoring.

Problem Statement

• ✅ Production-grade RAG system exists (src/rag/)
• ✅ Comprehensive experience data (736 lines, 15+ employers)
• ⚠️ n8n workflows use keyword matching (not using FAISS)
• ⚠️ No schema enforcement or guardrails
• ⚠️ ~5% hallucination rate
• ⚠️ ~45s end-to-end latency

Solution

Implement a 3-tier enhancement plan:

Tier A (1-2 weeks): Connect to FAISS + Add Schema

• Phase A1: Connect n8n to FAISS vector store
• Phase A2: Add JSON schema enforcement + guardrails
• Phase A3: End-to-end testing and optimization
• Expected Impact: 40-50% quality improvement

Tier B (3-4 weeks, optional): Fine-tune LoRA Adapter

• Collect SFT pairs from past tailored resumes
• Train LoRA adapter on Llama 3.1-8B
• Deploy and A/B test
• Expected Impact: 20-30% additional improvement + 2x faster inference

Tier C (ongoing): Advanced Features

• Prompt caching for repeated jobs
• Tool-use / function calling
• Feedback loop for continuous improvement

Success Metrics

| Metric | Current | Target (Tier A) | Target (Tier B) |
|--------|---------|-----------------|-----------------||
| Retrieval Precision@5 | N/A | > 0.8 | > 0.85 |
| Schema Compliance | ~70% | 100% | 100% |
| Hallucination Rate | ~5% | < 1% | < 0.5% |
| End-to-End Latency | ~45s | < 30s | < 15s |
| User Quality Rating | N/A | 3.5/5 | 4.5/5 |

Documentation

Comprehensive documentation has been created in n8n/docs/:

• README_ENHANCEMENT_PLAN.md - Index and navigation
• ENHANCEMENT_SUMMARY.md - Executive summary
• CODEBASE_REVIEW_AND_INTEGRATION_PLAN.md - Detailed analysis
• ENHANCEMENT_ROADMAP.md - Full roadmap
• QUICK_START_TIER_A_PHASE_1.md - Implementation guide
• TIER_A_PHASE_1_SETUP.md - Detailed Phase 1 guide
• TIER_A_PHASE_2_SCHEMA.md - Detailed Phase 2 guide
• IMPLEMENTATION_CHECKLIST.md - Day-by-day checklist

Related Issues

• Phase 1: RAG Integration - Index and Retrieve Experience Snippets #46: Tier A Phase 1 - Connect n8n to FAISS
• Phase 2: LoRA Fine-Tuning - Train on Curated Resume Tailoring Pairs #47: Tier A Phase 2 - Schema Enforcement + Guardrails
• Phase 3: Critique/Verifier Agent - Truth Filters and Self-Critique Loop #48: Tier A Phase 3 - End-to-End Testing & Optimization
• Phase 4: DPO Preference Optimization - Rank and Learn from Resume Variants #49: Tier B - LoRA Fine-tuning (optional)
• Phase 5: Metrics & Monitoring - Track Resume Quality Indicators #50: Tier C - Advanced Features (optional)

Files to Create/Modify

New Files (Tier A)

• n8n/scripts/query_faiss_vector_store.py - FAISS bridge
• n8n/scripts/validate_resume_output.py - Schema validator
• n8n/schemas/resume_output.json - Output schema
• .agent/prompts/system.tailor.guardrails.md - Enhanced prompt

Modified Files (Tier A)

• n8n/n8n/workflows/ingest.json - Add FAISS retrieval
• n8n/n8n/workflows/tailor.json - Add validation

Acceptance Criteria

• All sub-issues (Tier A Phase 1, 2, 3) are completed
• All success metrics for Tier A are met
• Documentation is updated
• End-to-end testing passes on 10 diverse job descriptions
• Quality metrics are documented

Labels

• enhancement
• rag
• n8n
• quality
• high-priority