Embeddings were a mistake. We matched OpenAI’s best embedding model on FiQA without using a single embedding. No vector database. No chunking. No re-indexing. Today, we're launching @wondersear
The Hook
“(No spoken dialogue — vintage projector click and ticking ambient score)”
Problem Agitation
“(No spoken dialogue — pulsing synth drone)”
Problem Agitation
“(No spoken dialogue — rising cosmic synth riser)”
Product Reveal
“(No spoken dialogue — UI mechanical typing clicks and warp whoosh)”
Feature Teaser
“(No spoken dialogue — gentle electronic confirm chime)”
Feature Teaser
“(No spoken dialogue — percussive electronic pulse)”
Call to Action
“(No spoken dialogue — emotional synthetic chords)”
Call to Action
“(No spoken dialogue — crystalline resonant logo sound effect)”
Synthesizable in Remotion · key components of the vintage-to-vector-knowledge-search template.
Full-frame historical monochrome footage overlaid with bold white year titles and yellow typewriter-style notes.
Stack Remotion Video components with CSS grayscale and film grain filters, layered under spring-animated bold SVG typography and yellow text spans.
Spiral vortex composed of thousands of text strings and glowing golden particles orbiting a bright central core.
Use Three.js canvas or Remotion Particle array with logarithmic spiral coordinates, animating rotation angle with interpolate(frame) and depth z-shifts.
User question translates into yellow predicted legacy database search tokens before triggering a light-speed tunnel pass.
Two text tiers using layout animations: interpolate opacities and translateY, followed by camera zoom blur via CSS backdrop-filter.
Two tall vertical columns formed of horizontal pixel segments that tick upward to target benchmark scores.
Render an array of div segments whose opacity/color triggers sequentially using frame-based intervals to build the 54.88 and 55.00 heights.
Purpose. Establish that companies lose vital tribal knowledge as generations leave.
Execution. 1950s factory footage paired with handwritten records and year stamps.
Purpose. Frame decades of organizational data as an incomprehensible cosmic cloud.
Execution. Spinning galaxy formed entirely of text snippets and luminous star dust.
Purpose. Demonstrate WonderSearch bridging natural language to legacy documentation.
Execution. Search bar query automatically converting into technical legacy search terms.
Purpose. Prove parity with state-of-the-art embedding models without setup friction.
Execution. Side-by-side segmented benchmark graphs showing matching scores.
Purpose. Leave an emotional and practical impression before company call-to-action.
Execution. Portraits of past workers transitioning into metallic glowing insignia.