Shipping multilingual content is rarely ‘translate and done’. Pages change, UI space is tight, legal terms must be exact, and markets never sleep. In The Agile Localization Podcast, host Stefan Huyghe talks with Putri Kumala (AI Localization Operations Team Lead) and Arshaad Mohiadeen (Senior AI Engineer) from Deriv about how they run 20+ languages with automated, AI-driven workflows while staying in control of quality, brand, and compliance.
Listen to the new episode on:
From ‘Who Will Translate?’ to ‘How Do We Automate?’
Putri says the biggest change is in mindset. Before, the team asked who would translate and how long it would take. After their shift toward full automation, the question became: how can we automate this so humans focus on judgment, not repetitive effort? That shift was scary at first for people who came up through translation. But once the team agreed that AI is a speed engine, and humans provide direction, context, and guardrails, the fear turned into a practical partnership.
"At the beginning, it was a little bit scary because I thought, okay, AI is doing translation for us, so what are we going to do here as a human? But as soon as we got together and started working with AI, it helped us a lot more in terms of speed.
The Pull and Push Workflow Everyone Understands
Deriv keeps the workflow explainable internally. Content is pulled from Webflow into Crowdin, pretranslation runs automatically, and translations are pushed back into Webflow. Putri says they keep the language simple: pulling and pushing. That matters because localization touches product, content, and engineering, and the system only scales if nontechnical stakeholders can follow what’s happening without a tooling deep dive.
Automation Doesn’t Remove QA, It Moves QA Up the Stack
A trading platform can’t afford sloppy wording, broken links, or confusing onboarding. So Deriv doesn’t stop the pipeline to translate, but they do add layers after the automated pass. Once content is pushed back, they run another QA layer and use AI for proofreading. They look for mistranslations, context misses, and functional issues, such as ‘Spanish’ link routing to the wrong language page.
Market volatility doesn’t usually change the workflow, Arshaad notes, because most content stays stable. The tricky part is the new jargon. When a term hasn’t appeared before, the team decides whether to keep it in English, translate it, or use transliteration, then updates terminology so automation stays consistent. Read more about localization testing.
Trust Is Measured Per Language, Not Assumed
Models aren’t trained equally across languages, and inconsistency is the real enemy at scale. Deriv maintains internal datasets and benchmarks multiple models against them, testing different prompt styles, translation output, and proofreading.
They also stress repeatability: ask the same sentence many times and watch for drift or hallucinations, because even tiny inconsistencies become a huge problem across 20+ languages.
"A good indicator for detecting hallucinations is to ask the same sentence or context multiple times. If you ask the same thing a thousand times, you want the AI to give something consistent, not random translations every hundred or thousand times.
Redundancy and Accountability Are the Scaling Strategy
Deriv runs 24/7, and providers can go down. Arshaad describes building redundancy: multiple providers and fallback options so workflows keep moving. But the key safeguard is accountability. When Stefan asks who owns quality, Putri’s answer is simple: the team does. AI can translate and proofread, but humans remain responsible for outcomes and for improving terminology, prompts, and rules over time.
Final Thoughts
Deriv’s playbook is less ‘use AI’ and more ‘engineer trust’. Automate the routing work, add QA layers that catch meaningful risk, benchmark models per language, and build redundancy so operations don’t stall. Then keep humans on the hook for quality, because trust isn’t a feature you ship once. It’s a discipline you should maintain every day.
Putri and Arshaad’s Background
Putri Kumala is the AI Localization Operations Team Lead at Deriv. With two decades of experience in localization, Putri has pioneered the transition from traditional human-led translation workflows to AI-first localization strategies while maintaining rigorous compliance and quality standards. Her expertise lies in balancing automation with human oversight, developing terminology frameworks, and integrating localization early in product design cycles.
Arshaad Mohiadeen is a Senior AI Engineer at Deriv, specializing in large language model implementation and multilingual AI workflows. With deep expertise in prompt engineering, model benchmarking, and quality assurance systems, Arshaad bridges the gap between AI capabilities and localization requirements in highly regulated environments. His work focuses on designing redundant systems, testing AI consistency across languages, and establishing frameworks that protect against hallucinations and model drift.
Listen to the new episode on:
Yuliia Makarenko
Yuliia Makarenko is a marketing specialist with over a decade of experience, and she’s all about creating content that readers will love. She’s a pro at using her skills in SEO, research, and data analysis to write useful content. When she’s not diving into content creation, you can find her reading a good thriller, practicing some yoga, or simply enjoying playtime with her little one.