v0.6.5 — H1B Data Page & Portal-Grade Job Filters

What's New

H-1B Intelligence

• New H1B Data page with FY2025 applications table (238K+ records)
• Lottery statistics across 5 fiscal years
• Country-cap status with cap-type indicators
• Human-readable company/city/state normalization in API responses

Job Portal

• 10+ filter dimensions on Jobs page: workplace type, employment type, degree level, max experience slider, ATS multi-select, visa sponsorship, date range, salary range
• Debounced real-time search
• Auto-discovery endpoint surfaces results when zero matches found

Frontend

• Full Neo Brutalist redesign with white/orange theme
• 12 page components (Home, Companies, CompanyDetail, Jobs, H1B Data, Offers, Compare, Predictions, LotteryCalculator, Terms, Privacy, NotFound)
• Version bumped to v0.6.5

Fixes

• CORS debug mode fixed for local development
• Resume match scoring hardened
• Timezone handling corrected (Django timezone vs datetime)

Full Changelog: v0.6.4...v0.6.5

v0.6.2

Fixed

• invalidated stale temporary resume-match sessions after matcher updates so old cached garbage results no longer survive page reloads
• fixed contact-line filtering so legitimate skills rows are kept during PDF parsing
• tightened the tailored PDF to prefer resume-backed matched skills instead of loose job-only phrases
• kept the Resume-Matcher-inspired whole-word keyword alignment while preventing old pre-fix sessions from rendering broken output

Verified

• docker compose exec backend python manage.py check
• docker compose exec frontend npm run build
• direct resume-matching verification against the provided PDF with fresh session creation, stale-session invalidation, and tailored PDF download generation

v0.6.1

Fixed

• kept temporary resume shortlist sessions across page reloads by storing only the session id in browser tab storage while still deleting raw PDFs immediately after extraction
• cleaned resume parsing so duplicated contact header noise no longer leaks into emails, skill chips, or match reasons
• improved resume skill extraction and keyword alignment to surface more believable high-match evidence like Python, React, TypeScript, and SQL
• narrowed role-family inference to phrase-level matches so false positives from substring hits no longer distort scoring

Verified

• docker compose exec backend python manage.py check
• docker compose exec frontend npm run build
• direct resume-matching verification against a real uploaded PDF with temporary session creation, processing, and tailored-PDF readiness

v0.6.0

Added

• Added a temporary resume-matching flow on the Jobs page that accepts a PDF upload, scores only the strongest-fit live roles, and generates one shared tailored resume PDF for the best-matching role cluster.
• Added public Jobs resume endpoints for upload, polling, session clearing, and temporary PDF download without creating long-lived candidate records.

Changed

• Reworked the Jobs page to include a dedicated resume studio with privacy messaging, processing state, shortlist results, and download actions while keeping the existing Ghosted visual theme.
• Updated the resume-matching pipeline to use live job-description similarity, role-family alignment, and years-of-experience gating instead of depending on a hand-maintained skill map.

Fixed

• Removed raw resume-file handoff between containers by switching the upload pipeline to in-memory extraction, so temporary sessions no longer fail when the worker processes the job on a different container filesystem.
• Tightened high-match filtering so the shortlist excludes more low-signal roles and finishes faster under the current ATS dataset.

v0.5.6

Highlights

• Removed the duplicate Companies top stats strip so the directory page no longer repeats the same headline metrics already shown on the homepage.
• Rebalanced the Companies results into a cleaner two-column desktop grid so 10-item pages land evenly instead of ending with an awkward orphan card.
• Kept the Companies page responsive, with no horizontal overflow on mobile after the layout change.

Verification

• docker compose exec frontend npm run build
• docker compose exec backend python manage.py check
• Browser validation on /companies for the cleaned header, even first-page grid, and mobile overflow safety.

v0.5.5

Highlights

• Added full keyboard navigation to the shared custom select so list filters work properly with Arrow keys, Home/End, Enter/Space, and Escape.
• Cleaned up the site navigation semantics and removed the duplicate hidden page heading so each route has a clearer accessibility structure.
• Added real labels to the Jobs search and location inputs and fixed Compare picker dismissal on Escape and outside click.

Verification

• docker compose exec frontend npm run build
• docker compose exec backend python manage.py check
• Browser validation on Jobs and Compare for keyboard selection, single-page heading structure, and picker dismissal behavior.

v0.5.4

Changed

• Switched the Dockerized backend runtime to PostgreSQL plus Gunicorn so the app no longer falls back to SQLite inside the container stack.
• Documented the Postgres-backed Compose behavior in the README and synced the tracked release metadata.

Fixed

• Added cached paginator counts for the filtered Companies, Offers, and Jobs endpoints so repeated page navigation no longer pays the same count cost on every request.
• Tightened the Jobs list queryset, added supporting database indexes, and widened long company slugs so the existing seeded dataset loads cleanly into PostgreSQL.
• Improved next and previous page prefetching on the main paginated pages so client-side navigation feels quicker once a page has settled.

Verification

• docker compose exec backend python manage.py check
• docker compose exec frontend npm run build
• API page-size checks confirmed 10 items per page on Companies, Offers, and Jobs
• Cold page-2 latency improved to about 170 ms on Jobs, 258 ms on Companies, and 173 ms on Offers under the Postgres-backed stack

v0.5.3

Changed

• standardized Companies, Offers, and Jobs to 10 items per page

Fixed

• added next-page and previous-page prefetching plus non-blocking page transitions on the main paginated pages so Next and Previous feel much faster once the current page settles
• optimized the Companies directory query by replacing heavier related-count joins with company-level subqueries
• trimmed the Jobs list query and replaced the expensive salary-record join count with a cheaper company-level subquery to reduce page-fetch overhead while preserving ranking signals

Verification

• docker compose exec frontend npm run build
• docker compose exec backend python manage.py check
• API verification confirmed page 1 returns 10 results for Companies, Offers, and Jobs
• browser verification confirmed warmed pagination transitions on Companies and Jobs complete quickly after prefetch

v0.5.2

Fixed

• balanced the Jobs results grid by expanding the last card across the full row when a 15-item page ends on an odd card count
• replaced the old favicon with a simpler Ghosted monogram and refreshed the icon links so browsers pull the new asset instead of stale cached branding

Verification

• docker compose exec frontend npm run build
• docker compose exec backend python manage.py check
• browser verification on /jobs confirmed page 1 still shows 15 jobs, the final card fills the row, and the icon links resolve to /favicon.svg?v=0.5.2

v0.5.1

Fixed

• widened the Jobs filter controls and updated placeholder copy so the inputs and selects stay readable instead of collapsing into tiny labels
• reworked Jobs result cards to give titles and metadata more breathing room, simplify match reasons, and improve scanability at common desktop widths
• preserved the Ghosted theme while keeping the Jobs page responsive with no horizontal overflow on mobile or desktop

Verification

• docker compose exec frontend npm run build
• docker compose exec backend python manage.py check
• browser verification on /jobs at desktop and mobile widths