🌟Vision & Mission

Sig Dug (founder) built Pickatale on a simple but powerful belief: children learn to read by reading things they love — not by grinding through worksheets.

How the loop works

Key principle

Reinforcement is invisible to the child. They pick a story about dinosaurs because they love dinosaurs. The system has already woven in the vocabulary targets, text structures, and comprehension challenges their curriculum requires — at exactly their reading level.

Market position

Who we serve

• B2B: Schools, school districts, ministries of education, publishers
• B2C: Families who want to support their child's reading at home

Library

~3,000 books from Oxford University Press, Arcturus, Cricket and others. Growing to 9,000+ through the Pickatale catalog API. Covers ages 4–11, multiple languages, all Flesch-Kincaid levels.

Competitive advantages

• Per-child AI agent (Learner Bot) — not just recommender algorithms
• Cross-nation analytics — unique dataset spanning multiple national curricula
• Full curriculum mapping (CM) — not just "reading stage" levels
• Adaptive content that re-levels and translates existing books, not just suggests new ones
• Modular, licensable components — each service can be sold independently

🎓Educational Foundation

Every product decision at Pickatale is grounded in reading science and child development research. This page summarises the key findings that drive our design choices.

1. Reading surface & typography

Background color

The reading surface must be calm. Vibrant design belongs in the library and rewards — not on the page the child is reading.

• Light mode (dark text on light background) improves visual acuity, reading speed, and comprehension across all ages.
• Warm backgrounds (peach, cream, warm white) further improve readability vs cool white — CMU research confirms shorter reading times.
• Dark mode drops comprehension by ~14% when contrast is poor or fonts are thin on pure black (halation effect).
• Cool saturated backgrounds (blue, green, purple) reduce readability — confirmed in multiple studies.

Sources: BOIA.org, tenacity.io, CMU research, UXMatters, teachertoolkit.co.uk

Typography for children

2. Finger-follow reading

Children drag a finger across text while reading. The platform highlights the word they're touching and counts those words toward their Miles total.

• Establishes left-to-right directionality and one-to-one word correspondence
• Creates multi-sensory learning (visual + tactile + kinesthetic) — strengthens memory pathways (University of Sydney, 2021)
• Maria Montessori advocated this technique from the early 1900s
• Develops phonemic awareness — children segment words into sounds more easily while tracking
• Beneficial for dyslexic readers — addresses visual processing challenges
• Finger position = proxy for attention focus — enables rich diagnostic telemetry

Sources: reading.com, thereadingspecialists.org, University of Sydney 2021, CEUR-WS.org

3. TTS with word highlighting ("karaoke reading")

The platform reads aloud while highlighting each word in sync — like karaoke. The child can follow along or read silently.

• TTS + synchronized highlighting = higher comprehension vs silent reading (phonics.org, ReadingRockets)
• Reduces cognitive load of decoding → frees attention for comprehension
• Improves reading rate, prosody, and syntactic consciousness (ResearchGate 2024)
• Less intimidating than reading aloud → motivates struggling readers
• Significant improvements in oral reading fluency among less fluent readers

Sources: phonics.org, readingrockets.org, ATIA.org, proquest.com, ODU research

4. Speedread (RSVP)

One word at a time, at configurable WPM — rapid serial visual presentation. Optional feature, off by default.

• Ages 8+ only — young children lack the working memory to process words fast enough for comprehension
• Max 10 minutes — visual fatigue sets in quickly
• Not the default — opt-in for motivated older readers
• Effective for fluent readers who want to increase their reading rate, not for building comprehension

5. Miles & motivation

Children earn Miles for words read (100 words = 1 Mile). Miles unlock more books. Tokens are earned per 10 Miles.

• Effort recognition (Miles) = pedagogically sound — rewards sustained engagement, not performance
• Tokens must unlock books, not prizes — extrinsic rewards for reading can undermine intrinsic motivation if they signal "reading is a means to an end"
• No leaderboards — competition creates anxiety in struggling readers
• No token loss — punitive mechanics harm motivation in young children

Sources: Self-Determination Theory (Deci & Ryan), reading motivation research

📖Reading Science

Understanding how children learn to read — and why it's hard — is foundational to every design decision we make.

The Simple View of Reading

Reading comprehension = Word Recognition × Language Comprehension. Both must be non-zero. A child who can decode perfectly but has no vocabulary won't understand what they read. A child with rich vocabulary but poor decoding can't access the text.

Pickatale addresses both:

• Word recognition: Finger-follow, TTS with karaoke, phonics-aligned decodables
• Language comprehension: Vocabulary taps, comprehension quizzes, curriculum-mapped content

Flesch-Kincaid (FK) levels

We use Flesch-Kincaid Grade Level as our primary text complexity measure. FK is computed from average sentence length and average syllables per word.

Zone of Proximal Development (ZPD) offset

The Adaptive Content Engine applies a +0.5 FK offset when serving leveled text. A child placed at FK 3.5 receives text leveled at FK 4.0. This ensures the child is always reading in their ZPD — slightly challenged, but not overwhelmed. Vygotsky (1978).

CEFR language levels

For English as an Additional Language (EAL) contexts, we map against CEFR:

Vocabulary development

The vocabulary gap is one of the strongest predictors of reading comprehension. Children who tap a word in the reader get an immediate popup definition. Those taps are stored as vocabulary gap data and fed to the Learner Bot — enabling the bot to target books that reinforce the specific words the child is struggling with.

Research-backed vocabulary targets per short non-fiction book:

• Ages 5–7: 3–5 target words (introduced before first encounter, with picture support)
• Ages 8–11: 4–7 target words (context-rich definitions, reused across pages)

📊Assessment & Leveling

Pickatale uses a multi-signal approach to understand each child's reading level — then continuously updates it as they read and take quizzes.

Placement assessment

When a child first joins, they complete a short placement assessment that establishes their baseline FK level. This becomes their fk_level in the database and drives all content serving decisions.

Continuous reassessment

Every quiz result updates the child's profile. The Learner Bot evaluates:

• Score > 80%: The child is likely ready to move up ("raise" signal)
• Score 65–80%: Stay at current level, serve more of the same
• Score < 65%: Consider moving down, flag for teacher attention

Quiz design

Quizzes are generated by the Adaptive Assessment Bot specifically for each book and FK level. They test:

• Literal recall: What happened? Who did X?
• Vocabulary in context: What does this word mean here?
• Inference: Why did X happen? (FK 4.0+)
• Text structure: What's the main idea? What's the author's purpose? (FK 5.0+)

Slow page detection

The telemetry system tracks time spent on each page. If a child spends more than 2× the average time on a page, it's flagged as a "slow page". More than 3 slow pages per session triggers an engagement_signal: struggling in the Learner Bot's memory — which the bot uses when selecting the next book.

🏗️Architecture

Pickatale is built as a set of independently deployable microservices, each with its own API and database. Services communicate exclusively via REST APIs — never via direct database access.

The four core systems

Data flows

Reading session → intelligence

Quiz result → bot update

Vocabulary tap → gap tracking

Infrastructure

• Server: 4 vCPU, 15 GB RAM (AWS EC2) — readingtester.com
• Reverse proxy: Caddy — automatic HTTPS, subdomain routing
• Databases: Single shared MySQL 8 Docker container (shared-db, port 3316). Each project gets its own database and user.
• Containers: All services run in Docker (docker compose per project)
• Internal networking: Services communicate via 172.17.0.1 (Docker host bridge)
• Auth: X-Internal-Key header for service-to-service calls. JWT for user sessions.

⚙️Services & APIs

Each service is independently deployable with its own versioned REST API. Any service can be licensed to external partners.

Intelligence layer

User-facing products

API conventions

• All APIs versioned: /api/v1/...
• External access: API key in Authorization: Bearer <key> header
• Internal calls: X-Internal-Key: pickatale-internal-key-2026 header
• Webhooks: every service supports webhook callbacks for key events (objective mastered, report generated, session ended)
• Response format: JSON, snake_case keys

🤖Learner Bot

The Learner Bot is the core of the Pickatale platform. Every child gets their own AI agent with a persistent memory store. It runs nightly, detects gaps, selects content, and prepares reports.

What the bot stores

Nightly run cycle

1. Pull all memory signals from the child's RAG store
2. Identify which curriculum objectives need reinforcement (gaps, low scores, slow pages)
3. Select or request generation of a book that targets those objectives at the child's FK level
4. Evaluate mastery from recent quiz results — update curriculum_state accordingly
5. Prepare reports: teacher (technical, with reasoning) + parent (warm, no jargon)

API endpoints

📚Content Pipeline

Content flows from curriculum objectives through to leveled, translated, narrated books served to each child. Every step is automated; the teacher's role is review and approval.

The three-step pipeline

Book structure

Short curriculum-aligned non-fiction books follow a research-backed structure:

1. Cover — title, illustration, grade level indicator
2. Learning goal — what the child will know after reading
3. Key words — 3–7 target vocabulary items with simple definitions
4. Content spreads (3–5 pages) — one idea per spread, visual + text
5. Did You Know? — surprising fact to hook interest and aid retention
6. Real World Connection — links the topic to the child's lived experience
7. Recap — retrieval practice: "Can you remember…?"
8. Reflect — higher-order question: "Why do you think…?"
9. Teacher layer (not shown to child) — curriculum objective, assessment guidance, differentiation notes

Adaptive Content Engine

The engine takes any book and:

• Re-levels text to any Flesch-Kincaid grade (GPT-powered, provider-swappable)
• Translates to any language (same call — one LLM does leveling + translation)
• Re-narrates with the appropriate voice style for the age group
• Applies ZPD offset (+0.5 FK by default)
• Caches the result — generate once, serve forever, no repeat LLM calls

Library vs Lesson Books

👩‍🏫Teacher Experience

Teachers are power users of the platform. They get full transparency into what the AI is doing — and always have the ability to override it.

Teacher portal

The teacher portal (teacher.readingtester.com) gives teachers:

Dashboard design principles

• Light theme — scannable, data-dense, less clicks (not dark)
• Dense information — class selector on top, everything visible without scrolling
• Big fonts — teachers read this on classroom displays, not just laptops
• No jargon — "Grade 3.5" not "FK 3.5" in the UI
• Override always available — teachers can always countermand the bot

What teachers see per student

📈Data & Analytics

Pickatale generates a unique cross-national dataset. No other ed-tech platform can compare reading outcomes across countries on equivalent curriculum objectives. The CM is territory-agnostic by design — any country, any curriculum framework, any language. Entering a new market is a content task, not an engineering task.

Analytics hierarchy

Privacy model

• Individual data: visible only to the child's teacher and parent
• Class/school data: aggregated + anonymized
• Nation/global data: k-anonymity enforced, never re-identifiable
• GDPR compliant — full data export and deletion available
• COPPA compliant — parental consent required for children under 13

xAPI / LRS

All reading events are logged as xAPI statements to our Learning Record Store (LRS). This enables interoperability with any LMS that supports xAPI (Tin Can API).

• Events: page_turned, word_tapped, quiz_completed, book_finished, session_started/ended
• All events include: learner_id, book_id, timestamp, FK level at time of event
• Offline events queued in Service Worker IndexedDB, replayed on reconnect

Offline-first telemetry

The Reader app works fully offline. The Service Worker intercepts telemetry POSTs when the device is offline and stores them in IndexedDB (7-day retention). Browser automatically replays them on reconnect — even if the app is closed. Each event has a UUID (event_id) for server-side deduplication.

🗺️Roadmap

What's built, what's in progress, and what's next for the Pickatale platform.

✅ Complete

• Curriculum Mapper (CM) — territory-agnostic, any country supported; Turkey, UK, Norway loaded as live examples
• Reader PWA — FK-leveled text, finger-follow, TTS karaoke, quizzes, vocabulary taps, Miles/Tokens
• Learner Bot — per-child RAG memory, quiz result processing, vocab gap tracking, session summaries
• Telemetry service — xAPI LRS, session tracking, offline-first Service Worker queue
• Adaptive Content Engine — FK leveling, ZPD offset, text caching (1,031 books × 3 levels)
• Teacher Portal — class dashboard, attention list, vocab gap view, student detail cards
• All 5 services connected end-to-end with verified data flows

🔨 In progress

• Lesson book generator (Step 2 of content pipeline) — teacher review + approve workflow
• Reader PWA deployment to app.readingtester.com
• Cross-grade book generation for CM objectives

📋 Next to build

• AI Orchestrator — pipeline manager connecting CM → Create → Bot. The missing piece.
• Pickatale Create API — generate custom illustrated books tied to specific objectives
• Cross-nation analytics — school_id, district_id, nation_code in learner onboarding
• Parent portal — warm reports, "Sofia had a great day!", no jargon
• Mobile narration pipeline — ElevenLabs/Fish Audio voice generation per book per language
• Curriculum objective → book mapping — populate curriculum_mappings table