cv-optimizer-feature

CV Optimizer Feature

Standalone Agents (NOT in LangGraph workflow)

Three agents in agents/cv_optimizer_loop.py plus HiringManagerAgent in agents/hiring_manager.py run on demand after a workflow session reaches completed. They are orchestrated by CVOptimizationOrchestrator — never add them to job_application_workflow.py.

Hiring Manager (evaluate) → convergence check → CV Optimizer (revise) → repeat
→ Cover Letter Finalizer → optimized CV + cover letter + gap analysis

Loop stops when: score ≥ threshold, max iterations reached, score decreases vs best, plateau detected, or API quota/rate limit (stop_reason: api_rate_limit, status: partial — saves best progress so far; cover letter skipped if quota hit before it runs).

Initial CV is composed from the user profile via _compose_cv_from_profile(). Job description comes from job_input_data["job_input"] with fallback to _synthesize_jd_from_analysis(job_analysis) when raw paste text is missing.

Background Task with Atomic Redis Locking

from utils.cache import (
    set_cv_optimization_running,   # Returns False if already locked (409)
    clear_cv_optimization_running,
    cache_cv_optimization,
    get_cached_cv_optimization,    # Unwraps cache_set {"data": ...} wrapper
    invalidate_cv_optimization,
    is_cv_optimization_running,
)

claimed = await set_cv_optimization_running(session_id)
if not claimed:
    raise APIError(ErrorCode.RESOURCE_CONFLICT, "CV optimization is already running...", status_code=409)

background_tasks.add_task(_run_cv_optimization_background, session_id=..., user_id=str(user_id), ...)

Always clear the lock in finally and broadcast lifecycle events:

from api.websocket import (
    broadcast_cv_optimization_started,
    broadcast_cv_optimization_iteration,
    broadcast_cv_optimization_complete,
    broadcast_cv_optimization_error,
)

Background task uses get_session(), not get_database(). Top-level except must call await report_exception(exc, user_id=...) and log with exc_info=True.