Emaki 絵巻 — kanji as pictures, with a GPT-5.6 tutor

Company: Emaki 絵巻 — kanji as pictures, with a GPT-5.6 tutor

Tagline: Kanji as pictures: scan real-world Japanese, learn with an honest GPT-5.6 tutor, and chain kanji like domino tiles. Built for Vietnamese learners, in the open with Codex.

Self-reported basis: The entire analysis is based on a single author-supplied description, submitted to the OpenAI 2026 hackathon on Devpost. No third-party verification or archived evidence exists.

Commercial due-diligence read: This is a self-directed educational tool built by one developer for personal use and demonstration. It claims to use AI to teach kanji through visual recognition and linguistic connections, particularly for Vietnamese learners. The product is not evidenced to have revenue, customers, or traction. The author's own account describes a prototype with limited functionality and no commercial deployment.

The description states that Emaki is an app that allows users to scan real-world Japanese text (e.g., signs, menus, manga) using their phone camera. It uses GPT-5.6 to recognize and explain each kanji, including its meaning, sound, and Hán Việt root if the user is Vietnamese. The app also includes a "domino tile" learning feature where mastering one kanji unlocks related ones. It supports offline use via a queue system and includes a game element called “Bông Ly,” a fox-powered duel engine.

The product is described as built with Codex, GPT-5.6, Claude, and other tools, and uses React, TypeScript, Tailwind, Vercel, and others for frontend and backend infrastructure.

Evidence: The author's own write-up

Confidence: Low — this is a self-reported prototype, not a deployed product with verified features or user behavior.

The author initially called the project "Kanji Domino" to emphasize the domino-like chaining of related kanji. It was later renamed "Emaki" (絵巻, a Japanese picture scroll) to reflect that it is not limited to one direction and can be used for learning across multiple language pairs.

The positioning is that this app leverages AI to teach kanji by connecting them to the user’s existing knowledge — especially relevant for Vietnamese learners who share roots with Japanese through Hán Việt. It also emphasizes honesty in AI use, with no fake AI or misleading outputs.

Evidence: The author's own write-up

Confidence: Low — this is a self-described evolution of intent, not validated traction or market positioning.

The app is explicitly built for Vietnamese learners of Japanese. It leverages the shared linguistic roots between Vietnamese and Japanese (Hán Việt) to make learning more intuitive.

It also implies broader use cases: “Japanese people learning Vietnamese or Chinese, other language pairs later.”

Evidence: The author's own write-up

Confidence: Low — no evidence of actual users or customer segments beyond the author’s personal experience.

Not evidenced. The description does not mention any pricing model, monetization strategy, or business model. It is described as a prototype built for demonstration and personal use.

Evidence: Not evidenced

Confidence: Very low — no indication of revenue, subscription, or commercialization.

The app is built with Codex (a terminal-based AI tool), GPT-5.6 family models (reasoning, vision, design), and Claude for review/testing. It uses React, TypeScript, Tailwind, Vercel, and other modern web stack components.

It includes features like:

• Vision proxy using GPT-5.6-luna
• Scan queue for offline use
• Game engine for domino-style learning
• Speech stack
• Rate limiting for public demo

The author also mentions 52 logged Codex work entries, with tests and failures documented.

Evidence: The author's own write-up

Confidence: Medium — the technical stack is described in detail, but no evidence of production deployment or scalability.

Not evidenced. There is no mention of users, customers, revenue, or adoption. It is described as a prototype built for a hackathon and personal use.

Evidence: Not evidenced

Confidence: Very low — no traction data or user feedback available.

Not evidenced. The description does not reference competitors, market size, or competitive positioning beyond the author’s own claims of lack of prior solutions that connect kanji to Hán Việt.

Evidence: Not evidenced

Confidence: Very low — no competitive landscape or market data provided.

• Unproven commercial viability: The app is described as a prototype, not a product with revenue or users.
• Single-person team: Only one developer (Hieu Phung) is mentioned, raising questions about scalability and long-term maintenance.
• AI dependency without fallbacks: While the app claims to be honest with AI use, it relies heavily on GPT-5.6 and other APIs that may not be stable or scalable.
• No monetization strategy: No evidence of how the product would generate revenue.
• Limited scope: The app is built for Vietnamese learners, which limits its potential market unless expanded.

Evidence: Author’s own write-up

Confidence: Medium — these are inferred risks from the lack of traction and commercialization.

1. What is your plan to scale beyond a single developer?
2. How do you intend to monetize this product, if at all?
3. Have you tested the app with actual users outside of your personal experience?
4. What are the technical limitations or scalability concerns with GPT-5.6 and Codex in production?
5. Are there any plans to expand beyond Vietnamese learners or Japanese text?
6. How do you plan to handle rate limits, API failures, and offline functionality at scale?

Evidence: Not evidenced

Confidence: Medium — these are necessary questions for due diligence but not directly supported by the description.