AI Assisted Portfolio Curation Workflow | Case Study

This is the portfolio follow-up system to Amir2000 Image Automation | Case Study. It uses existing AI quality values to surface strong candidates, then finalize selection manually and publish a curated portfolio page plus downloadable PDF.

0. Section Map (What Each Point Means)

• 1: Executive summary and explicit scope boundary.
• 2: Problem context and why this workflow exists.
• 3: Architecture and component responsibilities.
• 4: Picker logic and atomic file/FTP/DB flow.
• 5: PHP admin and public print/PDF behavior.
• 6: Full portfolio_v2 schema with origin markers.
• 7: Special cases and edge handling.
• 8: Operational outcomes and tradeoffs.
• 9: Key takeaway and artifact links.

1. Executive Summary And Scope Boundary

The portfolio workflow curates from a large existing archive into a tight public selection. AI assists candidate discovery and humans make final artistic decisions.

• Purpose: reduce manual curation effort while preserving creative/editorial control.
• Input: existing metadata and quality fields already stored upstream.
• Output: curated rows in portfolio_v2, website rendering, downloadable PDF.

Boundary clarification: this system does not generate metadata. Metadata generation happens in the automation pipeline before this stage.

2. Problem Context

Manual portfolio refresh does not scale well when the archive is large. Reviewing everything by hand every time is slow and inconsistent.

• Need a fast way to surface promising candidates from a high-volume catalog.
• Need deterministic publish assets (main JPG, thumb, WebP variants) for web and PDF usage.
• Need a workflow that keeps human artistic judgment as the final gate.

3. Architecture And Connected Systems

This case study is system 2 in a two-system chain:

• System 1: Amir2000 Image Automation | Case Study creates and stores metadata/quality fields.
• System 2 (this page): portfolio curation and publish shaping.

System 2 components:

• Standalone picker: portfolio_picker_v2.py (candidate browsing, add/remove/skip/special-entry).
• Admin page: portfolio_admin.php (visibility toggle, caption edits, drag reorder, PDF generation).
• Public print page: portfolio_print.php (ordered rendering, lightbox, open PDF link).

• Automation generates stable metadata and quality scores.
• Portfolio curation selects and packages the final photo set.
• SEO and performance engineering validates fast and efficient public delivery.

3A. System Diagram

The diagram shows the production chain from upstream automation to portfolio curation and final web delivery quality.

4. Portfolio Picker V2 Logic (Code-Level Workflow)

Picker behavior is explicit and transactional. For each selected image it creates exactly five versions and executes a rollback-safe pipeline.

• Five outputs per selected image: original archive copy (no watermark), web JPG 1500px, thumb JPG 600px, WebP main, WebP thumb.
• Watermark applies only to web and thumbnail variants and original archive copy stays untouched.
• Atomic sequence per image: pick -> copy original -> generate derivatives -> FTP bundle upload -> DB upsert.
• If copy/derivative/FTP/DB step fails: local backups restore and remote upload rollback is attempted before raising failure.

5. PHP Admin And Print Workflow (Code-Level)

Admin runtime (portfolio_admin.php):

• Session-based login with password hash verification against admin_users.
• AJAX endpoints: toggle visibility, save caption, save reorder.
• Filter controls: visibility set (portfolio/not/all), location, keyword, pagination.
• Reorder surface: SortableJS drag grid for currently visible portfolio rows.
• Embedded preview: includes portfolio_print.php and supports refresh without full reload.
• Client-side PDF generation: html2pdf snapshot from current reorder tiles and opens in a new tab for save/print.

Public print/runtime (portfolio_print.php):

• Reads visible portfolio rows ordered by Show_Order (then id) with a cap of 20 rows.
• Print setup uses 5 cards per row and 10 cards per page.
• Current setup shows 20 images total (two print pages).
• Renders EXIF summary + full-size links + Fancybox lightbox.
• Shows direct open button to static PDF (view or download): Amir_Darzi_www.amir2000.nl_image_portfolio.pdf.

6. Table Schema: portfolio_v2

The table below reflects the active schema used by this workflow and labels each column by runtime ownership/origin.

Legend: ** field value is inherited from photos_info_revamp for regular picker rows. *** field value is populated in the Special Add flow from EXIF, manual dialog input, or scoring.

Table constraints in current dump include a primary key on id and unique key (revamp_id, File_Name).

7. Special Cases And Edge Handling

• Special-entry rows are not linked to photos_info_revamp. They get revamp_id tokens like 1_001, 1_002.
• Special-entry rows are scored before insert by calling batch_image_quality_score.py through the sandbox scoring helper.
• If scoring paths are unavailable or scoring fails, special-entry insertion is stopped to avoid unscored rows.
• Picker upsert does not overwrite non-empty existing fields. It only fills missing values and protects Picked/Show_Portfolio/Show_Order/sort_order.
• Remove selected from picker deletes DB rows only. It does not delete already-generated local/remote image files.
• Skip/hide state is stored in a local SQLite file (tools/state.sqlite3), separate from MySQL.
• Admin rendering is currently capped to 20 visible rows in portfolio_print.php.
• Print setup uses 5 cards per row and 10 cards per page (2 rows).
• portfolio_admin.php controls Show_Portfolio and Show_Order. The picker intentionally does not control these workflow columns.

8. Operational Outcomes And Tradeoffs

• Strong candidate discovery is faster because ranking signals are already available.
• Final quality stays editorial because visibility/order are manually controlled in admin.
• Web delivery is consistent because each selected image is rebuilt into deterministic output variants.
• WebP variants reduce transfer weight and support the performance outcomes shown in amir2000.nl SEO and Performance Engineering | Case Study.
• Reliability is improved with per-image rollback logic for local files and remote FTP writes.

Tradeoff: human review takes more time per batch and it keeps portfolio quality and balance under manual control.

9. Key Takeaway

This workflow is a practical hybrid curation system: AI assists discovery and ranking, while human selection defines the final portfolio narrative.