About The Position
You will own the technical direction of OKX's next-generation social feed recommendation system — evolving it from a content feed into a unified recommendation engine that surfaces both content and platform features. Your decisions directly shape the experience of tens of millions of users and drive platform trading conversion.
Requirements
- Background — Master's or above in CS / Math from a top university; 8+ years of experience with 5+ years in core recommendation / search roles; track record of owning end-to-end recommendation pipelines at 10M+ DAU scale
- User Intent & Profiling (Core) — Experience designing unified intent representations across heterogeneous domains (content / feature / search); ability to fuse real-time behavioral signals with long-term stable preferences; hands-on experience with tiered user profile systems (cold-start → interest exploration → stable preference)
- Transformer & Ranking (Core) — Deep understanding of Attention mechanisms in sequential behavior modeling and their limitations (DIN / SIM / HSTU evolution); ability to propose independent solutions under engineering constraints; proficiency in Listwise losses (ListMLE / Softmax Loss) and joint multi-candidate ranking
- Multi-Task Training (Core) — Expert-level knowledge of MMoE / PLE / ESMM and gradient conflict identification and mitigation; ability to design composite loss function frameworks from scratch; proven methodology for bridging offline metrics (AUC / NDCG) and online business KPIs
- Business Attribution (Core) — Hands-on Uplift Modeling experience; proficiency in Position / Selection Bias correction and prediction probability Calibration
- Generative Recommendation (Strong Plus) — Understanding of Semantic Tokenization (FSQ / RQ-VAE) and conditional sequence generation; working-level knowledge of RLHF / DPO applied to recommendation systems
- Recommendation & Search Agent (Strong Plus) — Engineering experience with LLM Agent frameworks (Tool Use / ReAct); ability to define the collaboration boundary between Agent-based and traditional recommendation; experience designing systems that translate natural language intent into structured retrieval requests
- Engineering — Large-scale distributed training (10B+ parameter models); real-time feature engineering (Flink / Kafka); inference optimization under strict latency SLA
Nice To Haves
- First-author publication at RecSys / KDD / WWW ｜ Bandit / RL production deployment ｜ Background in fintech / crypto
Responsibilities
- Elevate the Ranking System — Drive continuous ranking model iteration with measurable impact on user retention and trading conversion
- Unify User Understanding — Build a cross-domain intent framework spanning content consumption, feature usage, and search, shifting the system from "what users clicked" to "what users are trying to do"
- Define the Technical Roadmap — Chart and execute a 12–24 month evolution from Transformer-based ranking toward generative recommendation (sequence generation + preference alignment)
- Pioneer the Agent Paradigm — Integrate recommendation and search capabilities into an LLM Agent framework, enabling proactive intent fulfillment rather than passive content delivery
Benefits
- Competitive total compensation package
- L&D programs and education subsidy for employees' growth and development
- Various team building programs and company events
- Wellness and meal allowance
- Comprehensive healthcare schemes for employees and dependants
- More that we love to tell you along the process!
Stand Out From the Crowd
Upload your resume and get instant feedback on how well it matches this job.
Upload and Match Resume
What This Job Offers
Job Type
Full-time
Career Level
Senior
