What DeepL Got Right That No LLM Can Copy

2026-05-15 · Source: Data Engineering on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

DeepL's translation product offers a real-time, word-by-word translation experience that surpasses general large language models (LLMs) and even Google Translate in quality and responsiveness. Unlike typical AI development workflows that prioritize LLMs and then address speed, DeepL focused on data first, selecting the appropriate model for specific speed requirements, and only integrating LLMs where they demonstrably improved the user experience. This approach allows DeepL to generate translations as the user types, adjusting dynamically without requiring a full sentence or explicit user action, creating a seamless and intuitive interaction that feels like co-thinking rather than tool usage.

Key takeaway

For product managers and AI engineers designing user-facing applications, prioritize user experience and responsiveness over a default reliance on large language models. Your team should evaluate specialized AI models and data-centric approaches first to meet critical speed requirements, reserving LLMs for tasks where their generative capabilities are truly indispensable, thereby avoiding fundamental design limitations that impact real-time interaction.

Key insights

Prioritizing data and speed over general LLMs enables superior real-time user experiences.

Principles

• User wait time dictates product feel.
• Match model to speed requirements.
• Data-first, then model selection.

Method

DeepL's method involves a data-first approach, selecting specialized AI models for specific speed needs, and integrating LLMs only when they enhance the core experience, ensuring real-time responsiveness.

In practice

• Optimize for real-time interaction.
• Consider specialized AI over general LLMs.
• Design for word-by-word output.

Topics

• DeepL
• Real-time Translation
• Large Language Models
• Specialized AI
• User Experience

Best for: Machine Learning Engineer, NLP Engineer, Product Manager, AI Product Manager, AI Engineer, AI Architect

Related on AIssential

• Fragments: June 16 — LLMs are transforming programming and business models, but require careful engineering and decentralized design to avoid pitfalls.
• Why Small Language Models (SLMs) Are Quietly Winning in 2026 — My Switch from Giant LLMs — Specialized small language models offer superior practicality, speed, and cost-efficiency over monolithic large models for most product use cases.
• I Taught My Local AI to Read Everything I’ve Ever Written — Local AI can privately learn from personal data using simple, self-hosted automation.
• The hard problems were never language problems. — Production AI agents require robust system architectures beyond just Large Language Models.
• LLMs just have more time and energy? — LLMs imitate human language patterns, including confidence and humility, through relentless, large-scale data compression, acting as a mirror of human communication.
• The ‘brownie recipe problem’: why LLMs must have fine-grained context to deliver real-time results — LLMs require fine-grained, real-time context and modular architectures to deliver effective, low-latency assistance in dynamic systems.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Data Engineering on Medium.