AI-powered speed reading engine with psycholinguistic optimisation.
Built by Ontos Labs
- RSVP Engine — Rapid Serial Visual Presentation eliminates eye movement bottleneck
- PDF Text Extraction — Upload any PDF, extract text client-side
- Information Density Weighting — Function words dim, content words brighten
- Word Frequency Timing — Rare words get more processing time
- Auto-Deceleration — Dense passages slow automatically
- Peripheral Preview — Next/previous word ghosts for lexical priming
- Context Ghost — Current sentence displayed for semantic framing
- Filler Compression — Strip function words for up to 40% compression
- Chunk Mode — 1-3 words per flash
- Focus Mode — Strip all UI, pure reading
- AI Paper Analysis — Claude identifies paper sections, generates summaries
- Section-Aware Pacing — Methods slow, discussion flows, abstracts fly
- AI Summary — Key findings distilled before you read
- Section Picker — Choose which sections to read
- Next.js 16 — App Router, React Server Components
- Vercel — Edge deployment, serverless functions
- pdf.js — Client-side PDF extraction
- Anthropic Claude — AI analysis (BYOK only, key never stored)
# Clone
git clone https://github.com/ash23x/neural-reader.git
cd neural-reader
# Install
npm install
# Dev
npm run dev
# Deploy to Vercel
vercel- Zero server-side costs on free tier (static pages)
- API route (
/api/analyze) proxies BYOK requests to Anthropic - User's API key travels: browser → serverless function → Anthropic → back
- Key is never stored, logged, or persisted
- Vercel Hobby tier = free hosting
MIT — Ontos Labs Ltd 2026