Typdit

Positioning

Typdit (“I typed it”) is a pen-name text community that treats the process of writing — not just the finished sentence — as the primary content. In an era when AI produces polished prose instantly, the trail of pauses, deletions, and revisions a human leaves behind has become more interesting than the result. Typdit makes that trail first-class.

Market and Problem

Social platforms keep optimizing for visual stimulation, performance, and curated personas. Users increasingly ask less “what did I see” and more “is this real?” Text-first communities partially counter this trend, but as long as only the finished post is exposed, AI-generated, plagiarized, and ghostwritten text remain visually indistinguishable.

Typdit’s bet on restoring trust runs through making the writing process visible. When pauses, deletions, and revisions appear alongside the final text, AI-generated text and human writing become visually distinct on the page — without requiring a classifier.

Core Audience and Personas

• Readers and writers seeking authenticity in the AI era — people interested in not just “skilled writing” but “how a sentence was actually arrived at”
• Pen-name short-form writers — those who want to escape the social self-censorship of real-name SNS but also avoid the irresponsibility of full anonymity → fixed pen-names
• Retro / monospace aesthetics audience — readers who feel an emotional affinity for terminal-style interfaces over decorated card UIs

Value Proposition and Differentiation

• Typing replay — A play button on every post lets readers re-watch how the writer arrived at their sentence. Pauses, deletions, and rewrites are visible. The reading experience becomes the path, not just the destination
• Fixed pen-names — accountable anonymity — Neither full anonymity (different ID per post, irresponsible) nor real-name (self-censoring). The same pen-name accrues posts and reputation over time
• Path-based post structure — Each post belongs to a context path (e.g., /philosophy/ontology/skepticism) shown next to the pen-name, so a writer’s posts cluster as a coherent thinking trajectory rather than disconnected moments
• Retro terminal UI — xterm-based monospace interface. Visual stimulation is intentionally suppressed so the writing itself sits in front

Core User Flow

• Enter: Auto-generated pen-name with no signup (or sign in to your existing pen-name)
• Browse: Scan the monospace feed by pen-name + path + first line
• Replay: Hit ▶ on a post to watch the typing process — see how the writer reached that sentence
• Write: Pick a path under your pen-name and type (keystroke events captured automatically so others can replay)
• Return: The same pen-name accrues writing — that is the identity. Follow / reputation are pen-name scoped

Writing → typing-replay exposure flow

flowchart LR
    Type["User typing
(keystroke event time series)"] --> Buf["Browser buffer"]
    Buf --> Submit["On publish: submit
final text + event log together"]
    Submit --> API["API server"]
    API --> Store[("MySQL
posts · keystroke_events")]
    Store --> Read["Other users browse"]
    Read --> Replay["▶ replay → re-render
event log client-side"]

Replay is not server-orchestrated — the keystroke time series travels with the post on publish, and the client replays it directly. No additional compute on read.

Business Model Hypothesis

Monetization is in the validation phase; current focus is accumulating user trust and a writing experience moat.

1. Pro pen-names (hypothesis) — Unlimited paths per pen-name, private paths, and writer-controlled replay visibility for a paid tier
2. Publishing / newsletter integration — Tooling to publish a pen-name’s accumulated writing externally (Brunch / Substack / newsletters). The “process metadata” itself becomes the differentiating content
3. AI-detection licensing — A keystroke-time-series classifier as a separate tool for publishers and educators to distinguish human writing from AI generation

System Architecture (planning decisions become system structure)

Two planning calls drove the system economics.

1. “Typing trails must replay without server cost.” → Keystroke events are not streamed to the server during typing. The client buffers them and posts the time series together with the final text on publish. Replay reads from that single payload — no per-replay server call. The cost profile of a Typdit post is essentially the same as a plain text post.

2. “Pen-names are accumulating identities, not single-use handles.” → Pen-names are stable identifiers under which paths, posts, and reputation accrue. Auto-generated pen-names can be re-acquired on next sign-in (device + pen-name token), so the writer can keep building under the same name without operational overhead.

Technology Choices and Trade-offs

• Frontend: Vite + React 19 + TypeScript + TanStack Query + react-router-dom + react-i18next + react-helmet-async + Tailwind CSS. xterm.js (@xterm/xterm + @xterm/addon-fit) powers the typing replay layer. The terminal aesthetic is not theming — it’s the natural visual idiom for keystroke replay
• Backend: Shared Hono API (api.typdit.com/v1/*), one core across nine services. Simple CRUD plus keystroke-time-series storage; a single warm process beats serverless on cost and complexity
• Database: MySQL 8.0 with a greenfield schema. Keystroke events stored as JSON time series — written once on publish, read once on replay. No index pressure
• Hosting: AWS S3 + CloudFront for the Vite build. SPA + react-helmet for share metadata. Static-hosting economics
• What was dropped: Image / video uploads (text-only is the differentiator), live streaming keystrokes (raises writer self-consciousness), algorithmic recommendation feed (path-based explicit navigation is the identity surface)

Operational Automation

• Pen-name auto-generation — Neutral Korean / English word combinations with collision-resolving suffixes. Users do not pick their own name (lower onboarding friction)
• Path normalization — Free-form input is normalized into a tree structure so posts cluster naturally
• Social OG composition — Every post auto-generates an OG image with pen-name + path + first line so the process metadata is visible the moment a link is shared
• Build guardrails — tsc + eslint + theme-lint enforce design tokens; new components stay consistent with the monospace aesthetic by construction

Current State and Operational Signals

• Status: Live. Pen-name accumulation, path tree, and typing replay are operational
• Started: 2026-02 (initial release on Vite + React, greenfield schema)
• Infrastructure: Single AWS Lightsail instance + S3 (typdit.com) + CloudFront, with the backend shared across nine services
• Verification signals: Pen-name-level return-visit patterns forming, average replay-per-post ratio actively monitored

Retrospective and Next Hypotheses

• What worked: Encoding “process over result” as a system feature (keystroke replay) rather than copy. Without the system primitive, the differentiator would have been invisible
• What I would redo: The initial monospace UI created an unfamiliarity barrier for some users. An onboarding moment that auto-plays a sample replay would likely have lifted first-post conversion
• Next hypotheses: (1) Pro pen-names — unlimited paths and writer controls, (2) Human-vs-AI classifier on keystroke time series, (3) External publishing / newsletter integration, (4) Pen-name reputation system

Comparable Engagements

The capabilities developed solo on Typdit transfer cleanly to other domains.

• Products where the process itself is the content — Anywhere AI is collapsing the value of finished output, exposing the human trail becomes the differentiator (writing, design, code, music)
• UX built on keystroke / interaction time series — Repurposing user behavior streams as content (auto-generated tutorials, learning analytics, review data)
• Pen-name / fixed-anonymous identity systems — Designing the middle ground between real-name and full-anon, with accumulated reputation
• Retro / monospace aesthetics as functional UI — Aesthetic decisions that are tied to a product’s core value, not just a theme
• Multi-service backend infrastructure for fast new launches — Adding a new product to a shared Hono core in a few days rather than weeks

I prefer engagements where one person carries the work end to end. Reach me via /work-with-me or /contact.