Resume Tailoring Phase 3: Feedback Loop Implementation Plan

Goal: Add approve/reject/edit/flag feedback on tailored resumes. Feedback is stored in MongoDB, embedded in Qdrant as voice examples or anti-patterns, and retrieved during future tailoring to improve output quality.

Architecture: New voice_feedback MongoDB collection stores feedback entries. On submit, approved/edited text is embedded via Voyage AI and upserted to Qdrant with chunk_type: "voice_example" or "anti_pattern". The existing tailoring pipeline’s retrieval step will pick these up automatically via source filtering. Frontend adds feedback buttons to the existing /admin/tailor page.

Tech Stack: Python, FastAPI, MongoDB (motor), Voyage AI (existing), Qdrant (existing), React, TypeScript, Next.js

Task 1: Voice feedback Pydantic model

Files:

• Create: backend/models/voice_feedback.py

Step 1: Write the model

"""Voice feedback models for resume tailoring quality loop."""
 
from datetime import datetime
from typing import Optional
 
from pydantic import BaseModel, Field
 
 
class VoiceFeedbackCreate(BaseModel):
    original_text: str = Field(..., min_length=1)
    final_text: Optional[str] = None
    feedback_type: str = Field(..., pattern="^(approved|rejected|edited|flagged)$")
    job_context: str = Field(..., min_length=1)
    note: Optional[str] = None
    section_type: Optional[str] = None
 
 
class VoiceFeedbackResponse(VoiceFeedbackCreate):
    id: str
    user_id: str
    qdrant_id: Optional[str] = None
    created_at: datetime

Step 2: Commit

git add backend/models/voice_feedback.py
git commit -m "feat: add voice feedback Pydantic model"

Task 2: Database collection getter and indexes

Files:

• Modify: backend/database.py

Step 1: Add collection getter

Add get_voice_feedback_collection async function following existing pattern.

Step 2: Add indexes in ensure_indexes

Add indexes on user_id, feedback_type, and job_context.

Step 3: Commit

git add backend/database.py
git commit -m "feat: add voice_feedback collection getter and indexes"

Task 3: Voice feedback handler — POST and GET

Files:

• Create: backend/handlers/voice_feedback.py
• Create: backend/tests/test_voice_feedback_api.py
• Modify: backend/app.py — register router
• Modify: backend/tests/conftest.py — register router in test_app

Step 1: Write failing tests

Tests for:

• POST /voice/feedback requires auth (401)
• POST /voice/feedback creates entry and returns it
• POST /voice/feedback with feedback_type “approved” triggers Qdrant embedding
• GET /voice/feedback requires auth (401)
• GET /voice/feedback returns user’s feedback entries

Step 2: Write the handler

• POST /voice/feedback — validates input, stores in MongoDB, embeds approved/edited text in Qdrant as voice_example, embeds rejected/flagged text as anti_pattern
• GET /voice/feedback — lists feedback for authenticated user, with optional feedback_type filter

Step 3: Register router in app.py and conftest.py

Step 4: Run tests, format, commit

Task 4: Frontend — feedback buttons on tailor page

Files:

• Create: frontend/src/pages/api/voice/feedback.ts — API proxy route
• Modify: frontend/src/shared/types/api.ts — add feedback types
• Modify: frontend/src/shared/lib/api-client.ts — add feedback methods
• Modify: frontend/src/pages/admin/tailor.tsx — add feedback UI

Step 1: Add API proxy route

Step 2: Add types and api-client methods

Step 3: Add feedback buttons to tailor page

After tailoring results are shown, display:

• “Approve” — saves all resume sections as approved voice examples
• “Reject” — saves output as anti-pattern
• “Flag” — opens note input, saves with note
• Success/error feedback after submission

Step 4: Commit

Task 5: Wire feedback into retrieval pipeline

Files:

• Modify: backend/services/tailoring_pipeline.py — retrieve voice examples alongside content chunks

Step 1: Update retrieval to include voice_feedback source

Add a second Qdrant search for source: "voice_feedback" chunks, filtered by user_id. Pass these to the generator as voice context.

Step 2: Update generator prompt to include voice examples

Add voice examples section to the prompt when available.

Step 3: Test and commit

Task 6: Format, full test suite, docs update

Step 1: make format Step 2: make test Step 3: make test-frontend-unit Step 4: Update _meta.ts with this plan Step 5: Commit