## The challenge

A top smartphone brand needed **real-time, on-device translation**—but traditional models **consumed too much RAM**, were **too slow for live conversations**, and couldn’t adapt efficiently across languages.

### Key Obstacles

- **Memory constraints:** High RAM usage limited multi-language support
- **Slow inference:** Laggy performance hurt real-time use cases
- **Complex customization:** No easy way to optimize for specific language pairs

## Our solution

We built **lightweight, language-specific models** using Liquid’s **customization stack** and **Edge SDK**, optimizing them for **low memory** use and **high speed** on smartphones.

## The results

The brand launched **fast, accurate translation** across devices—even with limited RAM.

- Dramatically reduced memory usage
- Real-time translation speeds for live conversations
- Custom-optimized models for priority languages
- Non-ML teams could deploy models easily