TypeForge is an adaptive typing trainer designed to help users improve speed and accuracy through targeted diagnostics instead of repetitive generic drills. Most typing tests measure a single WPM number and send the user back into another identical test. TypeForge instead tracks every keystroke, identifies exactly which keys and character transitions cause the most errors, and generates custom practice drills around those specific weaknesses using deterministic algorithms, not machine learning. Every drill is the output of rule-based logic analyzing the user's own keystroke history.
The interface draws inspiration from Apple's Human Interface Guidelines, emphasizing clarity, consistency, and focused interactions across typing modes, analytics, settings, and profile pages.
Video Demo
[フレーム]
Live Application
https://typingforge-six.vercel.app/
X Post
[フレーム]
Code
https://github.com/Type-Forge/Type-Forge
Screenshots
Typing Forge
Typing Forge
Typing Forge
How It Works
Every typing session feeds into a diagnostic algorithm that powers the rest of the app.
# === DIAGNOSTIC DRILL ALGORITHM ===
# 1. Track errors per key -> populate a QWERTY mistake heatmap
# 2. Extract weak character transitions and key bigrams
# 3. Generate custom word pools targeting those specific weak keys
# 4. Score each run with a diagnostic grade: WPM * (ACCURACY / 100)
Drill Mode uses this algorithm directly. A user can select specific keys they struggle with, say A and M, choose a training duration, and generate a custom drill built almost entirely around those letters instead of random word practice.
YOLO Mode takes it further. It's an endless adaptive training mode that continuously recalculates in real time as the user types, shifting focus toward new weak points as old ones get resolved. The training evolves as the user improves instead of staying static.
Architecture
TypeForge follows a unidirectional state flow: keystrokes are captured, processed by a pure logic engine, and pushed into Zustand stores that the UI reacts to.
Architecture
Settings, battle progress, and session statistics live in their own dedicated Zustand stores (settings-store, battle-store, stats-store), keeping each concern isolated and the typing engine itself completely decoupled from React. The full breakdown, including the keystroke sequence lifecycle, is documented in the repository README.
Real Results
I use TypeForge myself almost every day. When I started building it, I was averaging around 28 WPM. Today, my highest recorded speed is 67 WPM. A lot of the features in the app came directly from trying to solve my own typing problems rather than guessing what users might want.
How I Built It
The app is built with Next.js, TypeScript, Tailwind CSS, Zustand for state management, and Motion for animations. Google's Antigravity was used extensively throughout development for implementation planning, architecture reviews, UI iteration, debugging, and feature design. It accelerated experimentation while allowing rapid refinement of both the typing algorithm and the user experience.
I also built custom canvas-based sparklines for the performance trend charts instead of pulling in a heavy graphing library, and a custom Web Audio API click-sound system for tactile typing feedback, both to keep load times fast.
Prize Category
TypeForge connects to Best Ode to Alan Turing through its underlying approach to mistakes, not through a literal codebreaking game. Every typing error is treated like noise in a signal: the diagnostic algorithm isolates exactly where that noise is coming from and corrects it through targeted, adaptive drills, the same style of pattern recognition and signal analysis that made Turing's codebreaking work so influential.
Like Turing's codebreaking work, TypeForge focuses on identifying patterns hidden inside noisy data and using those patterns to improve outcomes.
It also qualifies for Best Google AI Usage. While TypeForge itself runs on deterministic algorithms rather than AI, the entire development process, from initial architecture through final UI polish, was driven by Google's Antigravity coding agent.
What's Next
TypeForge is free to use, with no ads, subscriptions, or paywalls. Planned next steps include authentication, cloud-synced profiles, and leaderboards, but the focus for this submission was shipping a polished, working trainer first.