AI Assistant for Intento TMX
The objective was to design and integrate an AI assistant within Intento’s Translation Memory Exchange (TMX) interface to help enterprise localization teams edit, rewrite, and refine multilingual content with greater speed and consistency. Intento, a provider of enterprise-level machine translation solutions, offers TMX as a platform for managing large-scale translation memory across teams and languages. Prior to the assistant’s integration, users manually edited translation segments with little contextual support, leading to slow localization workflows and inconsistencies in tone and formality across different markets.
My responsibilities
Stakeholder Workshops:
Collaborated with Intento’s product and customer teams to understand the daily pain points faced by language editors.
User interviews (4 users)
Conducted sessions with professional translators and localization managers to understand the editing process and challenges in translation managment.
Business Canvas Workshop with Stakeholders & Engineers
We clarified:The user outcomes we aimed to achieve.Hypotheses about user behavior. Key metrics to track success (bounce rates, filtering accuracy).User benefits and potential challengesEngineering redlines
Ideation and Design
Key exercises included:
User flows to map how users would navigate the AI assistant.
Storyboarding to visualize the user journey, particularly focusing on editing flow and text processing.
Later we built a first prototype.
User testing
Tested with 6 users. Plus we analysed the data coming from the first release of the feature.
Pain points & solution
Pain
Editors lacked support for tone and style adaptation when reviewing segments.
Solution
Introduced custom user prompts and a prompt management system, for prompts like “Rewrite in formal style” to allow tailored rewrites.
Pain
No guidance on why certain rewrites were better.
Solution
Added “Why is this better?” explanations under AI suggestions to build user trust and learning.
Pain
QA reviewers needed clearer signals about translation quality and reliability.
Solution
Integrated fluency rating labels (e.g., “Moderate”, “OK”) sourced from machine translation providers.
Design & Prototyping
Built initial flows in Figma focusing on side-panel suggestion layout, inline action buttons, and prompt inputs.
KPIs
+48%
Reduction in segment edit time
+62%
Fewer manual rewrites across large batches
97%
User satisfaction for AI-aided edits (internal survey)
Chacho Herraiz, Senior Product Designer.
Contact me
Chacho Herraiz
igdh.creative@gmail.com
©Chacho Herraiz 2024