Next-Word Prediction App — Investor Pitch

Slide 1 — Why it matters

Predictive text saves keystrokes and time.
Our prototype suggests the next word instantly, trained on real-world text (blogs, news, Twitter).

• Business value: faster typing, fewer errors
  
• Technical value: simple, explainable language model

Slide 2 — Algorithm (plain-English)

N-gram language model (1-, 2-, 3-gram) with Stupid Backoff (α=0.4)

• Precomputed frequency tables for unigrams, bigrams, and trigrams
  
• Lookup: try trigram prefix → backoff to bigram → fallback to unigram
  
• Keep top-k per prefix → lightweight memory & fast response

Output: one best next-word (optionally show top-3 suggestions)

Slide 3 — The App (shinyapps.io)

• Input: short phrase (1–3 words)
  
• Output: 1 predicted word (and optional top-3)
  
• Hosted at: https://phuoc-fpt.shinyapps.io/NextWordApp/

How to use: 1. Type a phrase (e.g., “according to the”)
2. Click Predict
3. See the predicted next word

Slide 4 — Results & UX

• Works on 5+ phrases from Twitter/News (leave out last word)
  
• Responsive UI; robust to casual/noisy text
  
• Simple interface; no training needed for end users

Example phrases to try:
“the president of”, “thank you for”, “in the middle”, “I can’t wait”, “according to the”

Slide 5 — Roadmap & Ask

Next steps - Better smoothing (Kneser–Ney), personalization, multilingual
- Top-k candidates with probabilities; profanity filter
- Edge caching; usage analytics

Ask - Support to productize the prototype
- (Optional) Hiring: bring this from prototype → production