Blog & Engineering Notes
Things we built, measured, or got wrong. Written by the developer, dated, and updated in place when code or benchmarks change.
The Physics of Sound Masking: Why Pink and Brown Noise Block Workplace and Speech Distractions
The psychoacoustics of spectral density (1/f vs 1/f^2): why harsh white noise causes listener fatigue, while balanced pink and brown noise mask intelligible speech and acoustic room spikes.
Previous Articles
Brain.fm and Endel Alternatives: Free Procedural Soundscapes for Focus Without a Subscription
Tired of paying $60 to $90 a year for background study audio? Discover how procedural acoustic masking and the Web Audio API provide non-repeating focus soundscapes with zero logins, zero trackers, and zero paywalls.
CEFR Vocabulary Profiling: How to Measure Video Transcript Difficulty from A1 to C2
How frequency corpora and CEFR thresholds determine video difficulty. Learn how to profile subtitle transcripts in real time to find your comprehensible input zone (i+1) without outdated academic profilers.
How to Stand Out in OpenAI Applications: A Hiring Engineer's Guide to Cold Outreach and Interviews
Direct guidance from OpenAI engineer Wulfie Bain on cutting through 500,000 views: the 3-bullet cold message formula, structuring technical answers, and conversational behavioral delivery.
The Step-by-Step Guide to Sentence Mining: Creating Contextual Anki Cards from YouTube
Traditional sentence mining requires brittle multi-extension stacks like Yomitan, ShareX, and AnkiConnect. Learn how Nativra lets you mine target words, sentence context, and audio timestamps directly from YouTube into Anki TSV decks with one click.
Readlang and Lute Alternative: Free In-Browser Video and Transcript Immersion
Readlang caps multi-word phrase lookups on free tiers, while Lute requires complex local Python server configurations. Learn how Nativra provides a 100% free, zero-setup alternative for in-browser video and transcript immersion with unlimited phrase lookups and synchronized Read Along.
FluentU Alternative: Why Video Language Immersion Does Not Need a $30/Month Paywall
FluentU locks curated video clips and interactive subtitles behind a $29.99 monthly paywall. Learn how Nativra provides a 100% free, zero-signup alternative that turns any authentic YouTube video into a language immersion lesson with dual synchronized subtitles and Anki export.
Multi-Offer Compensation Modeling: How to Compare Equity, Bonuses, and Cost of Living
Static salary boards show median figures but cannot model conflicting job offers. Learn how to compare base salaries, equity vesting schedules, sign-on bonuses, and tax brackets using Ravina's free client-side compensation modeller.
How to Run AI Mock Interviews With the STAR Method (Using Your Own API Key)
Proprietary interview prep platforms charge $25 to $50 per session for AI coaching. Learn how the STAR method rubric works and how Ravina lets you practice behavioral mock interviews with your own OpenAI or Gemini API key for pennies.
Rezi Alternative: Free ATS Resume Builder Without Export Paywalls
Rezi locks clean PDF/DOCX downloads and AI generation behind credits and subscription paywalls. Ravina is a 100% free, zero-signup alternative providing unlimited ATS-compliant single-column exports, local diagnostics, and BYOK AI directly in your browser.
Resume Worded Alternative: Free ATS Bullet and Layout Checking
Resume Worded locks line-by-line bullet rewriting and full scoring behind a $49 monthly paywall. Ravina is a 100% free, zero-signup alternative that runs 13 deterministic ATS formatting and bullet syntax checks locally in your browser with unlimited scans.
The Unquantified Bullet Penalty: Eye-Tracking Data on Recruiter 6-Second Scans
Eye-tracking research proves initial recruiter resume scans follow an F-pattern lasting between 6 and 8 seconds. Experience bullets lacking leading numerals suffer an 80%+ skip rate. Here is the Action Verb + Technical Scope + Measured Outcome syntax formula enforced by Ravina's deterministic bullet checker.
Subconscious Subvocalization in Second-Language Reading: Why Text-Only Study Blocks Listening Speed
Reading foreign text without synchronized audio forces the brain into internal subvocalization capped at 80-110 words per minute, causing auditory buffer overflows during 180 wpm native speech. Here is how synchronized text illumination recalibrates speech perception.
Why LaTeX Resumes Fail ATS Parsers: The pdflatex Ligature and Virtual Font Pitfall
Default pdflatex builds convert character pairs like 'fi' and 'fl' into composite ligatures that legacy ATS extractors drop or turn into whitespace. Here is why 'efficient' becomes 'e cient' in recruiter databases, and how to fix your LaTeX preamble.
Why Bring-Your-Own-Key (BYOK) Architecture Protects Candidates from AI Training Loops
When candidates upload resumes to closed AI SaaS platforms, personal salary history and patent drafts often enter fine-tuning pipelines. Here is how direct client-side BYOK API calls enforce enterprise zero-retention policies, reduce audit costs to $0.003, and protect candidate privacy.
The "We Did" Penalty: How Enterprise Interviewers Detect Candidates Hiding Behind Team Accomplishments
In behavioral loops at Amazon, Google, and Meta, using the collective pronoun "we" is the fastest way to earn a weak hire signal. Here is how calibration rubrics detect ambiguity, the 15/15/50/20 STAR scoring formula, and how to rewrite answers with individual ownership.
How Ambient Decibels and HVAC Acoustics Correlate with Deep Work Dropout: Insights from 5,000+ European Study Desks
Crowdsourced acoustic data from 5,835 European university libraries reveals why continuous low-frequency HVAC drone (48-52 dB) causes faster cognitive dropout than intermittent human whispers, and how room geometry dictates study stamina.
The Sub-Second Subtitle Desync Trap: Why Language Learners Get Exhausted Watching Foreign Cinema
Human auditory perception and phoneme mapping fracture when captions drift by as little as 200 milliseconds. Here is why broadcast TV subtitle standards create cognitive fatigue, how frame rate mismatches cause drift, and how sub-second alignment unlocks true comprehensible input.
Why ATS PDF Parsers Break on Hyphens and Em-Dashes (and Drop Minus Signs in Equity Models)
Word processors silently convert standard ASCII hyphens into Unicode en-dashes and em-dashes. Here is how that corrupts date parsing in legacy ATS pipelines, drops negative numbers in equity models, and how to verify your PDF in five seconds.
The Invisible ATS Trap in AI-Generated Resumes: Letter-Spacing and Broken Parsing
Why ChatGPT, Claude, and Gemini produce PDF resumes that look visually pristine to humans but extract as disconnected glyphs in enterprise applicant tracking systems like Greenhouse and Workday.
What We Got Wrong in Our Own SEO and AI-Search Setup: What We Changed
A one-person product audited its own site and found cloaked pages, a sitemap that lied about dates, placeholder entity links, and 41 template pages with 150 words each. Here is the list, the reasoning, and the fixes.
How an ATS Actually Reads Your PDF (and Why Two Columns Break It)
A technical walk-through of PDF text extraction (content streams, reading order, and why a layout that looks fine to you arrives scrambled in the recruiter's database). With a five-second self-test.