KOTCHAKORN
Neural Thai Language Learning Studio
Learning Thai is a beast. The tonal system alone has five distinct pitches that completely change meaning — the word 'mai' can mean new, burn, silk, wood, or be a question particle depending on tone. Most language apps reduce this complexity to gamified flashcards. I wanted something that felt like a pro-grade IDE for linguistics.
Origin Story
Kotchakorn started as a messy folder of Thai language notes I kept during sessions with my teacher in Bangkok. Scribbles in notebooks, voice memos, iPhone screenshots of restaurant menus. The data was there but completely unstructured. I realized I was building a personal linguistic corpus — and that corpus deserved an engine.
How It's Built
I developed this with active, iterative feedback from my Thai teacher to ensure the linguistic logic actually holds up. Every transliteration rule, every tone-mark mapping, every conversation scenario was validated against real-world Thai usage — not just textbook grammar. The app went through five major architecture rewrites before the current build.
Engineering Notes
Getting the tones right
Thai is tonal, and a wrong tone is a wrong word. Every transliteration rule and tone-mark mapping was validated with my teacher against real usage, not just textbook grammar — the diacritics picker even narrows suggestions by consonant class and vowel length, which is where most learners (and most apps) get it wrong.
Why fully serverless
The whole thing runs on Cloud Functions, Firestore and Firebase Hosting so there is nothing to keep warm and nothing to pay for when idle. App Check via reCAPTCHA Enterprise gates every generation endpoint and database operation, so the Gemini calls behind the lessons can't be scraped or abused.
From notes to a system
Kotchakorn went through five architecture rewrites. The breakthrough was treating my messy corpus — notebooks, voice memos, menu photos — as structured input to a synthesis engine rather than content to be displayed, which is what turned a study folder into a tool other learners could use.
Gallery
Kotchakorn is the app I wished existed when I started learning Thai. It is not a replacement for a teacher — it is a synthesis engine that sits between you and the language, transforming raw input into structured, reviewable, drillable knowledge. The name itself is a Thai name meaning 'born of the lotus' — growth from murky beginnings.
What It Does
Linguistic Studio
Generate deep-dive lessons on any niche topic — ordering street food in Ari, negotiating at Chatuchak market, reading temple signs in Chiang Mai. Each lesson comes with instant phonetic transliteration, IPA notation, Markdown-rendered explanations, and contextual usage notes. The generation engine synthesises cultural context alongside pure vocabulary.
The Simulator
Real-time conversation practice with AI-powered personas. Choose from preset characters: Local Tour Guide who speaks slowly, Street Food Vendor with rapid-fire colloquial Thai, Strict Professor who corrects every tone error, or Bangkok Taxi Driver who uses heavy slang. You can also define entirely custom personas to simulate specific encounters you're preparing for — a job interview in Thai, ordering medication at a pharmacy, or explaining a visa issue at immigration.
Tone-Mark Integration
A dedicated diacritics picker built into every text input across the app. Handles Mid, High, Low, Falling, and Rising tones without fighting your keyboard layout. The picker is contextually aware — it suggests the most likely tone based on the consonant class and vowel length of what you're typing. This alone saves hours of frustration that other apps simply ignore.
3D Semantic Deck
A stacked card system with physics-based interactions powered by Framer Motion. The AI extracts key vocabulary tokens directly from your generated lessons, groups them by semantic field, and presents them as interactive 3D cards you can flip, stack, and drill. Cards track your recall accuracy over time and surface weak spots automatically.
Advanced Lexicon
A full-scale personal dictionary and note manager. Organize vocabulary into nested folders, add Markdown-formatted notes with example sentences, tag entries with tone class and consonant group, and bulk export your entire lexicon to .md files for local archiving or Anki import. The search is instant and supports Thai script, romanised, and phonetic queries.
Dynamic Theming
Six high-fidelity visual modes built on a custom glassmorphism CSS framework. Black Obsidian for late-night study, Cyber Magenta for focus sessions, Jade Temple for a calmer aesthetic. Each theme transforms every surface, shadow, and accent colour — not just a background swap. The glassmorphism system handles backdrop-filter layering across all modal and card components.
Where It Stands
Kotchakorn is live and is the tool I reach for in my own Thai study, refined alongside my teacher. The natural next steps are spaced-repetition scheduling on the 3D deck and a wider persona library — the architecture already treats lessons as structured data, so both are additive rather than rewrites.
The frontend is built with Next.js 14 and TypeScript, using Tailwind CSS for utility styling and Framer Motion for the 3D card physics and page transitions. The entire infrastructure runs serverless on Google Cloud Platform — Cloud Functions handle the AI generation endpoints, Firestore manages user data and lexicon persistence, and Firebase Hosting serves the static build. Security is multi-layered: Firebase Auth with Google Sign-In for identity, and App Check via reCAPTCHA Enterprise to protect every API route and database operation from abuse. The AI backbone connects to Gemini for lesson generation and conversation simulation.