Search Polish, GPU Recording, Dark Mode

Week 12

Major search infrastructure improvements with target-level rendering and inline editing, GPU-accelerated recording workflow with multi-snapshot support, and theme system refinements across the portfolio.

#Inspiration Index

Search now renders at the target level with site-level deduplication strategy, eliminating redundant results across the board. Pipeline review gained inline entity editing and three new panels: processing history with detailed artifact tracking, section tree visualization with capture labels, and summary view with Shiki JSON syntax highlighting.

Navigation flyout now includes a 500ms open delay to prevent accidental triggers. Filter embedding cache delivers zero-latency section relevance sorting by precomputing embeddings at query time. Onboarding flow simplified with streamlined interest selection and inline search launchpad.

#Autoscroll Recorder

Multi-snapshot support arrived with interval-based viewport capture system and thumbnail strip UI. Screenshots now appear as inline previews on run detail pages. Dark mode received FOUC prevention and comprehensive color calibration across success states and UI elements.

Device presets expanded with mobile-standard and desktop-standard configurations optimized for T4 VGX GPU at 2x capture scale. Distribution routing UI now supports per-output S3 destinations with InfoPopover help text and duplicate destination controls. Runs can now include notes for context tracking.

Chrome for Testing replaced the previous browser setup with configurable recording parameters including zoom level, UI offsets, and trim seconds. GPU encoder switched from H.264 to HEVC/H.265 (hevc_nvenc) with expanded quality flags and p6 preset. Backend migrated from Xvfb to GPU-backed Xorg on AL2023 ECS GPU AMI for hardware-accelerated encoding.

HEVC encoding with hardware acceleration cuts file sizes while maintaining quality at 2x capture scale.

GPU migration

Frame deduplication using perceptual dHash eliminates redundant frames during transcode. CDN cache purging via Bunny Cache Buster service ensures fresh playback after indexing. Run/job ID separation now cleanly distinguishes per-attempt identity from stable S3 identity.

#bhonda.com

SEO metadata improvements rolled out across all routes. Theme system overhauled with blocking script to prevent FOUC and Tailwind v4 dark mode integration. Topics browsing launched with /topics/:topic route, PostCard component, and tagging taxonomy.

Blog section reintroduced alongside People profiles. Code highlighting arrived via CodeBlock and InlineCode components with GFM markdown support. List component added for structured content. Shiplogs and projects migrated to fully-local file-based system.

#PostgreSDK

Upsert operations landed with sdk.table.upsert({ where, create, update }) syntax. Transaction support added via sdk.$transaction() for atomic multi-operation workflows. Delete operations split into $softDelete() and $hardDelete() methods for explicit control.

Stale file cleanup now prompts before deletion with --force flag for non-interactive environments. Pull operation automatically detects and removes orphaned output files.

#Bunny Cache Buster

Bunny Cache Buster service started to purge CDN cache remotely for new and updated content.