Our Projects
Our focus lies in developing initiatives that benefit Chinland and the Chin community, particularly in Machine Learning, Artificial Intelligence, and Technology.
Products
Things you can use today
Applications built for readers, speakers, and writers of Chin — on the web and on the phone in your pocket.
Chin Translator
Chin Literature Award 2026AI Translator designed to translate English into Chin and Chin into English for multiple domains. Presently in the proof-of-concept stage.
ChinSpeak
A voice-first mobile translator for short, natural conversations between Chin (Falam) and English. Out now on the App Store.
Chin AI
An AI-powered knowledge platform built on Large Language Models with a Chin interface, so people can ask questions and work with the AI in their own language. Launching at chinai.chat.
Language technology
What those products run on
Every product above sits on models and data we build ourselves for Falam Chin (cfm). For a language with little on the shelf, the corpus and the benchmarks matter as much as the models — they are the part nobody can obtain by calling somebody else's API.
Falam Text-to-Speech
Our own Falam voice models — male and female, alongside English — serving every spoken result in Chin Translator and Chin Voice Lab today through a public /tts endpoint. A second generation is in training.
Falam Speech-to-Text
Falam Chin speech recognition trained on roughly 195 hours of aligned Chin news and audiobook audio, across Whisper large-v3-turbo with LoRA, Meta MMS, and NeMo FastConformer. It hears every word ChinSpeak and Chin Voice Lab transcribe, dub, and subtitle.
The Falam (cfm) stack
Three layers, each feeding the one below it. The corpus is scored against frozen benchmarks, the engines are built on what survives, and the tools put both in front of someone typing Falam.
Layer 01
Corpus & evaluation
The part that cannot be bought — years of Falam text and audio, aligned, audited, and scored.
falam-corpus — In use
Corpus ingestion, metadata, and versioned releases for Falam Chin (Laizo).
falam-benchmarks — In use
Frozen evaluation sets and scorers — the yardstick every model is measured against.
falam-review — In use
Local web tool for native-speaker adjudication of generated and corrected text.
falam-audit — Work in progress
Corpus intake auditing: point it at real Falam text, get back reviewable findings.
falam-align — Work in progress
Quality assurance for English ↔ Falam parallel pairs before they reach training.
falam-speech — In use
Forced alignment of audiobook audio to its ebook text, producing training clips.
falam-pipeline — Work in progress
The one-command intake run: raw texts and parallel pairs in, releases out.
Layer 02
Text engines
The linguistic layer every Chin product calls, from Unicode cleanup to grammar.
falam-normalize — Beta
Canonical Unicode, punctuation, and spelling normalisation for Falam Chin.
falam-restore — Beta
Contextual diacritic restoration — turns user-typed Falam into canonical spelling.
falam-tokenize — Work in progress
Word, sentence, and document tokenisation for Falam Chin.
falam-lexicon — Work in progress
The lexical database for Falam Chin, built to be rebuilt as the corpus grows.
falam-spell — Work in progress
Spell checking and suggestion ranking derived entirely from the lexicon.
falam-grammar — Work in progress
Context-aware checking that catches what a spell checker cannot.
falam-langid — Work in progress
Language identification that keeps Burmese and neighbouring Chin languages out.
falam-embed — Work in progress
Falam Chin embeddings, in the two forms that downstream search and RAG need.
falam-ocr — Work in progress
OCR ingestion for scanned Falam documents — the way old books enter the corpus.
Layer 03
Tools you can install
Where the stack reaches a Falam speaker's own keyboard and word processor.
falam-keyboard — Prototype
Type Falam without hunting for diacritics; restoration happens as you write.
falam-libreoffice — Work in progress
LibreOffice spelling extension (.oxt) built straight from the lexicon.
falam-word-addin — Work in progress
Offline Falam spell checking and suggestions inside Microsoft Word.
falam-studio — Work in progress
Local-first writing application sitting over every engine above.
Pet project
Built for the joy of it
Not everything needs a roadmap.
We are a diverse group with expertise across Data Science, Machine Learning, Analytics, and Linguistics. If you have an exciting project in mind, we are ready to join forces. Let's make a positive impact together!