Introduction to HYPERCONNECT AI
HYPERCONNECT AI is an organization that delivers real business results through machine learning research. We redefine business problems as ML problems and focus on improving the core metrics of our products. We avoid simply applying external solutions; instead, we rethink the problem setup from the ground up and build the best choices ourselves. As a result, the research outcomes of HYPERCONNECT AI have led to real business impact, and a small number of these cases have been featured in the Earnings Letter of our parent company, Match Group. At HYPERCONNECT AI, research is a means, not an end, and the goal is clear business impact.
Examples of problems we are solving at HYPERCONNECT AI include:
- Recommendation model scaling: While scaling laws are widely accepted for LLMs, they do not hold the same way for recommendation models. We conduct research to build structures in which performance improves as model and data scale grow. We work with large models, including LLMs, while designing models with realistic constraints such as inference in mind.
- Recommendation model alignment: Every recommendation problem ultimately comes down to quantitatively measuring "how much will the business grow when this content is shown to this user." We research methods to post-process scaled recommendation models and align them with business goals, based on reinforcement learning, causal inference, game theory, and related fields.
- Automating moderation ML processes with multimodal LLMs: Across multiple modalities such as images, audio, and text, we conduct research to automate and advance steps in Trust & Safety (moderation) ML workflows that were previously performed by humans, including data labeling, model improvement, and evaluation, using LLMs. We focus on making LLMs accurately understand the company's internal data and policies, and on making their judgments reliable at the level of internal experts.
We also publish meaningful research outcomes externally as papers or open-source projects. We do not aim for paper publication as a goal, but when solving product problems we encounter aspects that are not sufficiently explained by existing research, and we organize these into research outputs ([reference]). To date, we have produced about 20 external research outcomes.
To solve business problems with AI, the infrastructure for training models must also be properly in place. At HYPERCONNECT AI, we have built and operate our own research GPU cluster so that ML Research Scientists can fully carry out model development and experimentation.
ML Research Scientist Role
An ML Research Scientist at HYPERCONNECT AI views product problems as research problems and solves them. They lead the entire process, from problem definition, hypothesis formation, model selection and design, and experimentation, to performance analysis, with the goal of ensuring research results lead to real improvements in product metrics. Rather than solving problems already defined externally, they redefine problems to fit HYPERCONNECT's data and constraints. After broadly reviewing various published approaches and existing research, they select or newly create solutions suited to the situation. The core of this role is finding meaningful signals in products through repeated experimentation and validation ([reference]).
While collaborating with various functions, including PMs and software engineers, research direction and key decisions are owned by the individual. Beyond improving the performance of a single model, we place importance on building structures that can generate greater impact over time. Data and backend engineering is carried out in collaboration with dedicated teams.
Requirements
- Experience publishing papers with theoretical contributions at top-tier AI/ML conferences (NeurIPS, ICLR, ICML, CVPR, KDD, etc.)
- Experience leading a research or engineering team, or experience growing team members (including experience as co-first author or corresponding author)
- Ability to use causal inference and other statistical methods to extract meaningful insights from data and apply them correctly to decision-making
- Extensive development experience outside AI/ML, including client (Android, iOS) and backend
- Business-level English proficiency sufficient to persuade stakeholders
Preferred Qualifications
- Experience as first author of a paper in the main track of at least one top-tier AI/ML conference (NeurIPS, ICLR, ICML, CVPR, KDD, etc.)
- Research experience in core technologies related to recommender systems, reinforcement learning, LLMs, and AI Agents
- Experience productizing and iteratively improving ML systems in user-facing services handling large-scale traffic
- Ability to lead technical discussions with global teams in English
Hiring Process
- Employment type: Full-time
- Hiring process: Document screening > Recruiter call > First interview > Second interview > Final acceptance (*The process may be added to or changed as needed.)
- Application documents: Freely formatted, detailed career-based resume in Korean or English (PDF)
#LI-HP1
If any false information is found in the submitted materials, or if there are legal disqualifications for employment, the hiring may be cancelled. Additional screening and document verification beyond the hiring process announced in advance may be conducted when necessary.
Applicants who are recipients of state veterans' benefits will be given preference in accordance with applicable laws. If you are eligible, please declare this when applying and submit supporting documents upon hiring.
If you apply for a position at HYPERCONNECT, this Privacy Policy applies to the handling of personal information: https://career.hyperconnect.com/privacy
#HPCNT