Data Science Capstone | Product pitch
November 2026
One short phrase in. One relevant next word out.
Powered by a compact R n-gram model and an interactive Shiny app.
People write messages, reports and posts all day. A useful keyboard should reduce typing effort without adding complexity.
Training opportunity: the provided SwiftKey English corpora cover conversational, professional and long-form writing. Each style teaches different word patterns.
Lowercase and tokenize the phrase.
Look for the last 3 words in a 4-gram table.
Try 3-, 2-, then 1-gram tables if needed.
Return the most frequent continuation.
Lightweight frequency tables are precomputed offline, so the deployed app never has to retrain during a user request.
Try it: Type a phrase → press Predict next word → see one suggestion.
Example buttons and a clear action make repeat testing easy.
✓ One-click, one-word prediction
✓ Fallback for unseen text
✓ Compact model; no per-request retraining
✓ No application-level storage of user text
01 Train on a representative SwiftKey sample
02 Benchmark on held-out phrases
03 Improve top-1 accuracy and speed
04 Iterate with real user feedback