Tsundoku

Overview

Despite abundance of available knowledge, most people struggle with reading comprehension, retention, and depth. Proximo Lab envisioned “Tsundoku” (Japanese for acquiring books and letting them pile up unread) as an AI service that augments the reading experience—making comprehension easier, retention stronger, and engagement deeper.

Challenge

Reading is cognitively demanding. Readers encounter unfamiliar concepts, lose track of narrative threads, fail to connect ideas across chapters, and forget most of what they read within days. Traditional reading aids (dictionaries, study guides) interrupt flow and require active effort. The challenge was creating an AI system that:

• Enhances comprehension without interrupting reading flow
• Builds durable knowledge retention
• Adapts to individual reading styles and knowledge levels
• Works across diverse content types (fiction, non-fiction, technical, literary)
• Respects reader autonomy (augments, doesn’t replace, thinking)

Solution

Tsundoku is an AI reading companion that operates in three layers:

1. Real-Time Augmentation

• Inline definitions and context for unfamiliar terms (personalized to user’s knowledge)
• Character and concept tracking across long texts (visual relationship maps)
• Thematic thread highlighting (identifies recurring ideas, motifs)
• Temporal and spatial grounding (maps, timelines for complex narratives)

2. Active Recall System

• Spaced repetition prompts surfacing key concepts days/weeks later
• AI-generated review questions testing understanding
• Concept linking exercises connecting current reading to previous books
• Knowledge graph visualization showing accumulated understanding

3. Deep Engagement Tools

• AI discussion partner for exploring ideas from the text
• Socratic questioning mode that challenges reader’s interpretations
• Comparison engine finding resonances with other works
• Critical analysis prompts developing analytical thinking

Technical Implementation

• Language Understanding: GPT-4 based comprehension models with custom fine-tuning on literary analysis
• Knowledge Graphs: Neo4j database tracking entities, concepts, and relationships across user’s reading history
• Personalization: User modeling tracking vocabulary, background knowledge, interests
• Spaced Repetition: Modified SM-2 algorithm for concept review scheduling
• Content Processing: NLP pipeline for entity extraction, theme identification, narrative structure analysis
• Multi-modal Interface: Web reader + mobile app + browser extension + Kindle integration

User Experience

Reading workflow:

1. User imports book (upload or link to source)
2. Tsundoku analyzes content, building concept map and identifying key themes
3. As user reads, AI provides contextual augmentation (hover/tap for definitions, concept cards)
4. AI tracks reading patterns, noting concepts that cause re-reading or slow-downs
5. After reading sessions, AI generates personalized review questions
6. Over days/weeks, spaced repetition system surfaces concepts for reinforcement
7. When reading new books, AI identifies connections to previous reading

Impact Metrics

After 6 months with 2,000+ users:

• 65% improvement in comprehension test scores vs. control group
• 3x better retention after 30 days (measured by quiz performance)
• Users complete 40% more books per year
• 85% of users report “significantly deeper engagement” with material
• Average session length: 45 minutes (suggests sustained engagement)

Key Innovation

Most reading apps focus on speed (speed reading) or convenience (text-to-speech). Tsundoku optimizes for depth—it assumes reading is valuable precisely because it’s cognitively demanding, and provides scaffolding that makes that challenge productive rather than frustrating.

The AI doesn’t summarize books or provide shortcuts. It helps readers build genuine understanding by:

• Making implicit concepts explicit
• Revealing structural patterns
• Creating memory anchors through active recall
• Facilitating connections across texts

Philosophical Approach

Tsundoku embodies a specific theory of learning: knowledge is networked, not isolated. Understanding a concept means connecting it to existing knowledge. The AI’s primary function is facilitating these connections—both within a text (how this chapter relates to earlier ones) and across texts (how this idea appears in other books you’ve read).

This positions AI not as a replacement for human cognition but as a cognitive tool—akin to how writing augments memory. Tsundoku augments comprehension and retention, enabling readers to engage with more complex material than they could manage unaided.

Commercial Model

• Freemium SaaS: Basic augmentation free, advanced features subscription
• B2B licensing: Educational institutions, corporate learning programs
• API: Developers integrating reading augmentation into their applications

Future Development

• Multi-language support (currently English-only)
• Academic paper mode (specialized for technical/scientific literature)
• Collaborative reading (shared annotations and discussion)
• Audio mode (augmented listening for podcasts/audiobooks)
• Expert knowledge integration (linking reading to structured knowledge bases like Wikipedia, academic databases)