Skip to content

AI Workflows

While language is the focus of the site, the underlying platform serves as an experimental playground for AI assisted development and workflows.

The Stack (Local-First)

The platform is designed to be high-performance, private, and extremely low-maintenance.

  • Frontend: Built on MkDocs Material (static site generation) heavily augmented with Vanilla Javascript.
  • Persistence: We use a Local-First architecture. Progress is instantly saved to your browser's localStorage ensuring sub-millisecond UI feedback. We integrate Userbase.com for optional end-to-end cloud sync, allowing you to back up your streak across devices.
  • Interactivity: Custom-built JS game engines, Orbit-powered Spaced Repetition (SRS) flashcards, NotebookLM authored podcasts and slides, and a Giscus-powered feedback system.

AI-Assisted, Human-Led Workflow

The core 10 lessons serve as the manually curated "Source of Truth," constructed after weeks of research. However, the interactive layers built on top of them are powered by AI:

  • Agentic Coding: The game engines, UI features, and cloud sync logic were co-developed alongside Antigravity (an agentic coding AI).
  • Audio Generation: We utilize Python scripts to interface with Google Cloud's advanced Text-to-Speech models (like Chirp3-HD) to generate the native-sounding, street-accurate pronunciation clips.
  • Content Expansion: Workflows utilizing Gemini Deep Research and NotebookLM help process grammar rules, generate podcast scripts, and maintain architectural consistency across the Markdown codebase.

Behind the Curtains

The project is entirely open-source (GitHub Public Repo). However, we maintain a private staging repository that acts as an internal "brain"—holding thousands of lines of raw research, Python build scripts, and experimental features before they are distilled and deployed to this live site.

Disclosure: I work at Google, but this is a pure passion project built on the side to keep myself accountable to my learning goals. The AI stack utilized throughout this project consists exclusively of Google's public tools (Gemini, NotebookLM, Google Cloud TTS). Keep exploring, keep practicing, and remember: Banni, swalpa swalpa kaliyona! (Come, let's learn little by little!)

Comments