Staff Software Engineer, Machine Learning

Our Mission:

As humans, there are few things more exciting than meeting someone new. At Tinder, we’re inspired by the challenge of keeping the magic of human connection alive. With tens of millions of users, hundreds of millions of downloads, 2+ billion swipes per day, 20+ million matches per day, and a presence in 190+ countries, our reach is expansive—and rapidly growing.

‍

We work together to solve complex problems. Behind the simplicity of every match, we think deeply about human relationships, behavioral science, network economics, AI and ML, online and real-world safety, cultural nuances, loneliness, love, sex, and more.

About the Team

Seoul is one of Tinder's key global hubs — home to cross-functional product pods building for hundreds of millions of members across 160+ countries. Our Seoul-based teams span Product, Design, Engineering (iOS, Android, web, and backend), and Data, working side by side to move fast: incubating new ideas, running rapid experiments, and shipping features that scale from a single market to Tinder's global business.
We partner closely with teams across the region and our U.S. headquarters, taking what works locally — new experiences, growth strategies, product bets — and turning it into playbooks and products that shape how Tinder shows up for members everywhere.

Team Introduction

The Tinder ML team drives impact across nearly every core domain of the product — Recommendations, Trust & Safety, Profile, Chat, Growth, and Revenue optimization. Our mission is to apply machine learning to enhance user experiences, foster trust, and accelerate business growth across Tinder’s ecosystem.

About The Role

We are looking for a Staff-level Machine Learning Engineer who can apply generative AI — large language models and vision-language models — across multiple Tinder product surfaces in service of real user problems. This is not a single-feature role and it is not a research role detached from the product. You will work across multiple product surfaces, decide where generative AI genuinely changes the user outcome and where it does not, and ship the ones that do.

This role owns Tinder's generative AI strategy and technical vision. You will define the multi-year architecture for how LLM-powered capabilities are built here. You will partner directly with the multiple stakeholders, ranging from Product, Design, and Engineering to translate that vision into a roadmap the organization can execute against, and you will be the person the company looks to for a clear-eyed read on what generative AI can and cannot do for dating in the next 1~2 years.

Just as importantly, this role is expected to drive its own project bottom-up. You will drive bottom-up initiatives from a blank page through production and measured impact, acting as the de facto product owner when no one else holds that ground. That demands a holistic view — of Tinder's product and its user base on one side, and of the model, data, evaluation, and serving stack on the other — and the judgment to see where a technical capability and an unmet user need meet.

Where you'll work:

This is a hybrid role and requires in-office collaboration three times per week. Our office is located in Seoul. 

In this role, you will:

• Conduct deep-dive research into large language models and vision-language models, and translate findings into concrete architectural decisions for Tinder's products.
• Own the generative AI technical strategy and vision, including the multi-year architectural direction for LLM-native systems and the build-versus-adopt decisions on models and infrastructure.
• Originate and drive bottom-up generative AI initiatives end to end, from problem framing and prototype through launch and measured impact.
• Design and ship generative AI capabilities across multiple product surfaces, including recommendations, profile, conversation, and Trust & Safety.
• Architect LLM-native systems that hold consistent behavior across large volumes of concurrent sessions while keeping per-session latency and token cost within production budgets.
• Build the evaluation infrastructure that makes generative quality measurable offline and online, so generative features can be iterated on with the same rigor as classical ML systems.
• Apply vision-language models to multimodal product and safety problems, including content understanding, ranking, and moderation.
• Diagnose and resolve failure modes in production LLM systems, turning each into a durable evaluation or architectural fix.
• Serve as a strategic thought partner to engineering managers and senior stakeholders on the state of the generative AI field and its product implications.

You’ll need:

• PhD in Computer Science, Machine Learning, or a related field, or an MS with an equivalent research record, with 5+ years of combined doctoral and industry machine learning experience.
• A sustained peer-reviewed publication record at top-tier venues (e.g., NeurIPS, ICML, ICLR, ACL, EMNLP, NAACL, CVPR, or equivalent), evidencing the ability to conduct original research
• 2+ years shipping LLM-powered systems in production at consumer scale, with end-to-end ownership from problem definition through launch and post-launch iteration, and accountability for quality, latency, and cost in a live product.
• Demonstrated experience applying generative AI to a large-scale consumer social, dating, or real-time interpersonal product.
• Working familiarity with the product dynamics of the dating domain, including reciprocal two-sided matching
• Demonstrated experience driving the strategy and vision for a new generative-AI-powered user experience in this domain, and securing organizational alignment behind that direction across technical and executive audiences.
• A track record of originating and leading ambiguous, bottom-up initiatives from a blank page to production impact, including operating as product owner when formal ownership is absent.
• Depth in LLM systems architecture, including context and memory design, retrieval, agentic tool use, persona consistency across long multi-turn interactions, and bounding inference cost at scale.
• Practical experience with post-training and alignment methods, and with the failure analysis needed to understand when and why models fail to follow specifications.
• Experience of building evaluation systems for generative outputs