Most search engines are simple. I decided to make this one ridiculously complex. It is built to filter just 1,000 static JSON entries, but it runs 21 parallel algorithms in real-time in your browser. It combines exact matches, fuzzy logic, semantics, and personalization. Yes, it is beautifully over-engineered.
The simplest and strongest signal. If the user types the exact full name of the target.
"github" → GitHub
Matches the start of a name. Rewards matches that cover more of the total word length.
"face" → Facebook
Matches the prefix of any individual word inside a multi-word target name.
"dri" → Google Drive
Uses Longest Common Substring logic to find continuous character blocks anywhere in the target.
"tube" → YouTube
A curated map of highly specific, manually verified abbreviations distinct from auto-generated acronyms.
"yt" → YouTube
Calculates spatial distance on a QWERTY keyboard. A slip of the finger (like pressing 's' instead of 'a') gets penalized less than a completely wrong key.
"amaxon" (x is near z) → Amazon
Breaks text into trigrams (3-letter chunks) and compares overlap. Brilliant at catching jumbled letters or missed spaces.
"linkeidn" → LinkedIn
Encodes words based on how they sound when spoken aloud in English. Catches massive spelling errors as long as they sound similar.
"fasebuk" → Facebook
Extracts the first letter of each word in a multi-word target to match against user initials.
"aws" → Amazon Web Services
Implements strict Damerau-Levenshtein transposition checks to perfectly identify when two adjacent characters are swapped.
"googel" → Google
Penalizes results if the user types multiple words (e.g., 'google docs') and the target only contains one of them, effectively burying mismatched pairs.
"google docs" ⊘ Google Maps
Uses an internal graph of related concepts. If you search for an intent, it finds the tool.
"email" → Gmail, Outlook
Runs standard match algorithms against the target's underlying description text instead of just its name.
"music" → Spotify (music streaming)
Splits the user query into individual words and ensures the target satisfies multiple concepts simultaneously.
"buy stocks" → Robinhood
Direct matching against hidden categorical tags assigned to the target data.
"chat" → Slack [tag: chat]
Strips out URL protocols (https, www) and explicitly scores against the root domain.
"figma.com" → Figma
Tracks global frequency. Queries that are searched often get a slight organic boost.
+ up to 15% bonus
Associates specific queries to specific clicks. If you always click 'GitHub' when searching 'g', GitHub learns to win.
+ up to 20% bonus
Maintains an LRU stack of your last 20 visited targets. Recently used items are strongly preferred.
+ up to 8% bonus
Dynamically shifts rankings based on the user's local clock. Work apps peak 9-5, media peaks in the evening, news peaks in the morning.
"docs" at 2PM → Google Docs
Tracks short-term query sequences. If you search 'google' and immediately search 'drive', the engine cross-references them to boost 'Google Drive'.
Contextual Memory
Intelligently bypasses expensive O(N²) fuzzy matching algorithms when scanning massive text fields (like 300+ character descriptions) in large datasets, instantly falling back to a blazing-fast O(N) boundary regex matcher to prevent UI blocking.
0.0ms Execution Time
Prevents memory leaks and garbage collection stutters by hoisting heavy Soundex phonetic dictionaries out of the hot-loop, stopping the browser from re-allocating memory 4,000 times per keystroke.
0.0mb Memory Leaks
N-Gram sets and Phonetic tokens are pre-computed at application load, guaranteeing zero allocation overhead during live search.
0.0ms Overhead
Prevents firing the search engine on every millisecond of a keypress sequence, instead running once when the user pauses typing.
150ms Buffer
Memorizes previous exact search queries and instantly serves the cached result set when a user hits backspace or re-types a query.
Instant Retrieval
Pre-allocates global Float32Arrays for Keyboard Distance and LCS algorithms, eliminating expensive garbage collection during typing.
O(1) Memory Setup
Stops searching for fuzzy duplicates the microsecond it fills the UI render limit, preventing O(N²) browser freezes.
O(N) Lazy Evaluator
Dynamically constrains the Substring matching algorithm based on word counts to prevent random syllables from inflating scores.
Strict Boundaries
Click the (i) button on any search result to visually reverse-engineer its exact mathematical breakdown and highlighted substring extractions.
Radical Transparency
A Stage-3 Re-ranking pipeline that prevents generic prefix queries (like 'goo') from flooding the top 5 with products from a single company.
Exploratory Routing