Fintech Algorithm Expert

Role Introduction

You will own a frontier algorithm topic and take it from research all the way to production — design, train, evaluate, deploy — and be accountable for its online performance. We don't give you a task list; we give you a sufficiently large problem and a rare dataset in the industry: multi-domain behavioral data across payment, mobility, food delivery, and credit, for the same cohort of users, spanning the full temporal scale from millisecond-level event tracking to daily / weekly-level transactions. The methodology is yours to define; we only care about how far you can push this problem.

Key Directions

You may own a topic (examples, not exhaustive). We want you to go deep in one direction and become its owner, rather than executing all tasks simultaneously (we're not looking for a generalist, but someone who can go deep in one direction, design experiments to reach reliable conclusions, and dare to bring frontier methods to production):

1. Financial Behavior Sequence Foundation Model — Pre-training objectives, tokenization architecture, scaling, multi-source data mixing ratios, Account-Card dual-dimension and cross-temporal-scale fusion are all open research questions on this line;

2. Representation Learning and Self-Supervised Learning — Contrastive learning, masked prediction, cross-modal fusion, enabling one set of representations to serve multiple downstream scenarios;

3. Identity Recognition and Biometric Verification — Model capabilities in remote identity verification, liveness detection, comparison, and anti-attack, supporting identity trust across the full chain of account opening and payment;

4. Customer Qualification and Fact Feature Mining — Not recording behavior, but inferring undisclosed objective facts about users from indirect signals — inferring relationship and gang association networks based on graph neural networks, mining macro fact features such as residence and business district based on remote sensing / geographic data;

5. Risk Control and Credit Modeling Deployment — Integrating frontier representations into card fraud detection, credit scoring, and other scenarios, designing blending and fine-tuning, and completing online deployment.

Qualifications

Required:

1. Master's degree or above in Computer Science / Mathematics / Statistics or related fields;

2. 3+ years of deep learning R&D experience, with experience building and completing the full model training pipeline from scratch (including pre-training or large-scale training);

3. Proficient in Transformer and its variants (GPT / BERT / FT-Transformer / MoE), with practical experience in sequence modeling or representation learning;

4. Familiar with mainstream deep learning frameworks (PyTorch preferred), with distributed training (multi-GPU / multi-node) experience;

5. Solid experimental design ability: able to independently conduct ablation studies, scaling law experiments, and reach reliable conclusions.

Preferred:

1. Experience in the financial domain (risk control, anti-fraud, credit scoring, etc.);

2. Foundation Model / self-supervised learning publications in top conferences (NeurIPS / ICML / ICLR / KDD / WWW, etc.);

3. Experience in contrastive learning, ELECTRA, cross-modal fusion, time series / event sequence modeling (TimeMixer / TrajGPT, etc.), graph neural networks / graph anomaly detection.