Yucheng Chen

Business Development Executive

Emeryville, California, United States4 yrs experience

Most Likely To Switch

Key Highlights

Developed a scalable quantitative research framework.
Achieved 19% annualized return in backtested strategy.
Ported operating system to RISC-V, winning academic recognition.

Stackforce AI infers this person is a Fintech and Software Development professional with strong quantitative and algorithmic skills.

Contact

Skills

Core Skills

Quantitative ResearchMachine LearningOperating SystemsCompiler DesignPerformance OptimizationSoftware DevelopmentAlgorithm DesignCompetitive Programming

Other Skills

LinuxC (Programming Language)GitC++Object-Oriented Programming (OOP)Python (Programming Language)RISC-VParallel ComputingAssembly LanguageGraph TheoryNumber TheorySearch(BFS, DFS, IDA...)Operations ResearchXAIFactor Filtering

Experience

China international capital corp

2 roles

Market Making Trader

Jul 2024 – Present · 1 yr 8 mos

Maintain daily market liqidity trading requirement. Responsible for L2 strategy development (spread strategy).

LinuxC (Programming Language)GitC++Object-Oriented Programming (OOP)Python (Programming Language)+22

Quantitative Strategy Developer

Jan 2023 – Jun 2023 · 5 mos · Beijing, China · On-site

I built a lightweight quantitative strategy research framework, which includes data processing, feature calculation, model training, and backtesting modules. The research framework has been adopted by the corresponding research team and has proven to be scalable and efficient in daily real-world applications.
Additionally, I carried out futures-spot arbitrage research on the framework. For strategy, I constructed 70 features including high-frequency factors and technical indicators. Ornstein–Uhlenbeck process is used to filter the data. Then used the Logistic Regression to predict the spot and future spread. When the cumulative income of the model prediction window exceeds 0.25%, then open a position. The position will be closed when the accumulated profit reaches the stop-profit stop-loss point. The strategy achieved an annualized rate of return of 19% in the 20-22 year backtest.
Design a research framework that consist of data processing, feature calculation, model training, and backtesting modules. Framework was implemented in python.
The data processing module allows users to define dispatch & reduce and file reader primitive to suit a wide range of use cases.
The feature calculation module use closed-form lambda functions to support the addition and modification of feature factor calculation. Calculation dependency is resolved by python runtime evaluation to reduce time & space overhead.
The model training module is designed with a pipeline structure and supports pytorch & sklearn machine learning algorithms. Multilabel learning is supported in framework.
Backtesting Module includes the strategy, the position, and the time modules. Each module is decoupled and allow users to self-define their strategy, position statistics and time frequency.
A future-spot arbitrage was conducted using the framework and show its efficiency in real-world application.

Python (Programming Language)LinuxQuantitative ResearchMachine Learning

Tsinghua university

Project Researcher - David Patterson RISC-V International Opensource Laboratory

Aug 2020 – Jun 2021 · 10 mos

1. Porting Harmony LiteOS-M Operating System to RISC-V Platform
Through kernel assembly transplantation, kernel code adaptation and driver code transplantation, Harmony LiteOS-M system is transplanted to RISC-V open sourceISA. This work is adopted by related chip companies and hardware teaching institutions. This work has won the outstanding undergraduate Thesis.
2. Compiler Target Specific Instruction Scheduling Leveraging LLVM TableGen
LLVM tablegen is used to generate a cost model of RISC-V Rugrats Chip and machine instruction scheduler is used to run heuristic on the cost model. Embench(www.embench.org) is run on the performance model of RISC-V rugrats to evaluate the effectiveness of the scheduling. Average of 8.5% cycle number reduction comparing with not scheduled code, and average of 3.8% cycle number reduction
comparing with the LLVM generic scheduler is achieved.

Assembly LanguageCompiler OptimizationOperating SystemsCompiler Design

Tencent

Software Engineer Intern

May 2020 – Aug 2020 · 3 mos · Guangzhou, Guangdong, China

Software infrastructure engineer in wechat group(WXG).
ANN is one of the important basic components supporting Tencent's WeChat recommendation service. Currently, HNSW (hierarchical NSW method) is an ANN algorithm with the best online query performance.
However, the HNSW solution adopted by Tencent cannot run all cores under high load conditions of multiple cores. I did a CPU hotspot analysis and a control comparison experiment on the corresponding industry source code library. The scalability bottleneck is located, that is, resource allocation and function binding cause excessive multi-threading overhead. I take a consistent hash map approach, tying resource allocations to threads. Successfully increased the peak throughput of HNSW multi-threading adopted by Tencent by 14%-16%, and the CPU multi-core efficiency increased by 8%-10% on average

LinuxGitSoftware DevelopmentPerformance Optimization

Sun yat-sen university

Team Member - Sun Yat-Sen University Algorithm Competition Representative Team

Sep 2017 – Dec 2018 · 1 yr 3 mos

I won the gold medal in the ACM-Internation Collegiate Programming Contest and got 410/500 points in China Computer Federation CSP certification (top 0.27%, 18th Nationwide). And served as the Guangdong Provincial Judge Committee of the NOIP.
The following types of topics are mainly dealt with in the competition:
DP: Graph DP, Resource DP, State Compression DP, Monotone Queue DP, and others.
Data Structures: Segment Tree, Index Array, Mergeable Heap, Splay, Persistent Data Structure
String: String Hash, Suffix Array, Suffix Automata, Palindrome Automata
Graph Algorithm: Shortest Path, Search, Heuristic Search, IDA*, Chain Decomposition, Node Divide & Conquer.
Also, combinatorics, number theory, probability, etc. topics are included.

Data StructuresC++Algorithm DesignCompetitive Programming