Typing on a phone is slow. A smart keyboard that predicts your next word — like SwiftKey’s — turns three or four taps into one.

• Trained on ~500,000 real sentences from blogs, news articles, and tweets
• Cleaned and standardized the text (removed links, usernames, profanity)
• Built a statistical language model that has learned which words tend to follow which

The model looks for the most specific match to what’s already been typed — and if it hasn’t seen that exact phrase before, it falls back to a shorter, more general one, so it always has an answer ready.

This is a live result from the trained model, not a mockup.

A web app: type a phrase, click Predict, get the top 3 next-word guesses — just like the suggestion bar on a phone keyboard.

• Try it now: nicsookraj.shinyapps.io/next-word-predictor
• Five example phrases built in, for a one-click test drive
• Fast: loads in ~1.3 seconds, predicts in a few milliseconds
• Lightweight: the entire trained model is ~11 MB

What’s working today:

• Deployed, tested end-to-end, and live for anyone to try
• Handles both everyday phrases and unfamiliar input gracefully — it never just gives up

Being upfront about a limitation:

• Right now, common short phrases can occasionally outrank a more specific longer match. It’s a known tuning opportunity, not a blocker.

What we’d build next:

• Train on the full text corpus, not just a 12% sample
• Refine the scoring so specific matches are favored more consistently
• Learn from an individual user’s own writing style over time