Promptly Translated

Main Screen

Inspiration

The inspiration came from my personal frustration with existing translation tools. As an indie game developer, I needed to translate my newsletter and game content, but popular translation platforms consistently produced awkward, robotic-sounding results. I noticed that while LLMs could generate beautiful, natural text in German, they struggled to translate to German effectively. This made me realize that translation isn’t just word-swapping—it’s about recreating the natural flow and cultural context of native speech.

What It Does

Promptly Translated introduces a multi-step, context-aware workflow:

Setup a project and provide background information, define tone, key terminology, and brand-specific translation rules
Input your text and select target languages
The input is sent to an LLM which identifies unclear or ambiguous passages and generates questions with suggested answers
The user can review and edit these questions and answers to provide better clarity
A first translation pass is executed, using all available context: project setup + clarified answers
The translated content is then refined by another LLM pass to ensure natural flow and native-sounding quality

Key features:

Supports text, HTML, and key-value pairs
Perfect formatting preservation
Batch translation into multiple languages
Terminology management & protected brand terms
Project-based customization with tone and style guidance
Runs entirely in your browser with your own OpenAI API key

How I Built It

Stack: React (Vite) + Tailwind CSS with a dark, tech-inspired theme
OpenAI integration: Direct API usage with advanced prompt engineering
Architecture highlights:
- HTML structure validation to prevent broken markup
- Robust error handling with automatic retries
Data storage: Everything stays local-only

Challenges I Ran Into

Token limits: Preventing content truncation while maintaining translation quality
- Solution: Dynamic token calculation, full context-window usage, clear user warnings
HTML translation: Preserving structure while translating content
- Solution: Sophisticated validation + retry mechanisms
Natural flow vs. literal accuracy: Extensive prompt engineering and a multi-pass process to balance both
Ambiguous source content: Creating an LLM-driven system to generate clarification questions and handle user-provided responses
Prompt design challenges: Crafting effective prompts for clarification, data structure extraction, translation, and review refinement

Accomplishments I’m Proud Of

Translations that sound native, not machine-generated (validated in German & Italian)
Handles complex HTML while keeping structure intact
Professional workflow support: terminology management, tone control & project organization
Intuitive UI—accessible to non-technical users yet powerful for experts
Question-answer system that actively improves translation quality and user control

What I Learned

Translation ≠ text generation—requires specialized techniques and user involvement
The context-first approach (project info + question clarification) yields superior results
The multi-pass translation process dramatically improves naturalness
Token management is crucial; users need feedback on content size limits
HTML translation demands a careful balance between content and structure
Users benefit greatly from being involved in clarifying ambiguous phrases before translation
Adding a structured clarification step helps the LLM focus and improves both quality and trust in the output

What’s Next for Promptly Translated

Let users create and edit their own clarification questions to refine meaning more precisely
Backend integration to enable persistent storage and offer a paid service model
Support for additional LLM providers (currently only OpenAI is supported)
Gather structured user feedback to drive ongoing improvement and usability
Long-term: explore integrations with dev tools, CMSs, and collaborative translation workflows