sync: Add role identification overlay on results page

sync: Switch role extraction to gemma3:4b instead of deepseek-v3.2

sync: Fix LLM role extraction: increase max_tokens 30 -> 500

sync: Distinguish AI vs keyword role extraction in progress messages

sync: Use LLM to extract ideal job role for MCP scraper

sync: Add MCP scraper source — LinkedIn, Indeed, ZipRecruiter, Glassdoor via Crawlee

sync: Remove JSearch and Apify — unused sources with no API keys configured

Cast Nango end_user.id to string (Nango rejects numeric ids)

Fix Nango connect/sessions 400: drop invalid callback_url, read data.connect_link

sync: Add Adzuna job search — 50+ countries, 150 results per search

sync: Replace Composio with Nango for LinkedIn integration

Fix pydantic version conflict — composio-core requires >=2.6.4

Sync backend: add Composio LinkedIn import + company enrichment

Simplify cover letter prompt to 4 paragraphs + relax Stripe webhook signature

feat: HN 'Who is Hiring?' scraper via Algolia API

feat: custom sign-in and sign-up pages with branded layout

fix: icon/placeholder overlap in all inputs + sidebar shows user name

fix: rename sidebar link 'Pro Plan' → 'Pricing'

feat: promo code field on subscribe page

feat: introductory pricing — $11.99 offer, $19.99 struck through

feat: subscription page + Pro Plan sidebar link

fix: show Load more jobs only on the last page

fix: remove broken inner import resume_converter in 3 endpoints

fix: correct seniority matrix — overqualified penalty for senior→junior

feat: quality gate, 5-dim scoring engine, improved semantic ranking, ATS selectors

feat: add description column to saved_jobs, store and return it

feat: LLM-powered ATS scoring via Ollama Cloud (llama3.3:70b)

fix: update root route message — frontend hosted on Vercel

deploy: backend restructure + dynamic Clerk auth + new job sources