Cvly · AI Career & Salary Intelligence
AI career intelligence and anti-fabrication resume tailoring platform indexing 6,577 live jobs across 7 Iranian boards.
01 — The Operational Problem & Stakes
Job seekers in Iran jump across fragmented boards (Jobeelo, JobVision, IranTalent, EEstekhdam, Quera, Karboom, Estekhdam24) sending identical generic resumes — yielding an estimated ~2.6% interview rate versus ~9.3% for tailored applications (industry estimates, Nov 2026).
At the same time, naive LLM resume rewriters hallucinate employers, dates, or skills the candidate never had, destroying credibility in technical interviews.
02 — System Architecture & Execution
Cvly (live on R23.2 @ `98a9ea0`, verified 2026-11-18) indexes 6,577 live listings (Nov 2026 snapshot) across 7 sources into a tuned SQLite WAL engine (`busy_timeout=30000`, shared inter-worker `PrivateTmp` caches, sub-second loopback response) behind FastAPI, RTL Jinja2, and HTMX/Alpine.
Users can run a zero-signup Instant Fit Check (`POST /api/instant-check`), explore 832 real salary benchmarks on `/salaries`, generate per-job tailored resumes guarded by a deterministic anti-fabrication validator ported from JobHunt, and export crisp Persian PDFs.
03 — Interactive Subsystem Topology (Click Any Node)
2-Hour Scheduled Board Crawlers
Systemd timer ingests, deduplicates, and normalizes 6,577 live listings with zero `mock_*` placeholder records.
04 — Key Engineering Decisions & Trade-Offs
Pivoting From Blind Mass-Apply to Targeted Fit Packages
Live PM review proved captcha-heavy boards block blind cloud form-filling and mass-apply hurts candidate conversion; focusing on instant fit scoring + verified resume tailoring solved the real bottleneck.
Fixing SQLite High-Load Lock Contention
Removing synchronous `backfill_job_roles` from worker startup and adding WAL + `busy_timeout=30000` + shared cache brought `/` down to `0.99s` loopback.
Zero-Signup Value First (`/api/instant-check`)
Visitors test a real job fit analysis right on `/` and `/score` before creating an account, seamlessly carrying state into onboarding.
Global LLM Semaphore & Daily Token Guard
Prevents upstream LLM latency spikes from exhausting FastAPI workers or exceeding daily cost ceilings.
05 — Verified Build, Test & Runtime Proof
$ curl -w "%{http_code} %{time_total}s\n" -o /dev/null https://cvly.ir/
200 1.18s (loopback: 0.99s | /score: 0.98s | /login: 0.97s)
✓ Live SQLite index: 6,577 active job listings across 7 boards (0 mock_* rows) — Nov 2026 snapshot
✓ Salary intelligence: 832 verified signals active on /salaries | registered users: 331+ (Nov 2026 snapshot)