Alexey is available for hire
Hire AlexeyAlexey Rodriguez Yakushev
Verified Expert in Engineering
Data Scientist and Developer
Location
Berlin, Germany
Toptal Member Since
July 1, 2022
Alexey是一位拥有13年以上机器学习经验的专业人士, data science, and software engineering. 他是大型推荐系统的主要贡献者, music and audio analysis, click-through rate prediction, RTB auction price prediction, computer vision, compilers, and tools for software parallelization. Alexey enjoys working with teams to build impactful products.
Portfolio
Deepspin
Python 3, PyTorch，数字信号处理，图像处理，Python...
Freelance
Spark, SQL, Google Cloud Platform (GCP)， Python 3, Machine Learning, Scala...
SoundCloud
Spark，机器学习，Python 3, Scala, SQL, C, c++， TensorFlow，数据科学...
Experience
Availability
Full-time
Preferred Environment
Linux, Python 3
The most amazing...
...我所做的事情是为超过1亿的用户提供每周推荐系统，支持分布式计算基础设施和低延迟服务.
Work Experience
Senior Machine Learning Engineer
2022 - 2023
Deepspin
- 实现了磁共振成像的几个图像重建算法. 该算法使用了从内部低成本MRI原型获得的MRI信号. 结果对于满足与项目资金相关的里程碑是至关重要的.
- Improved the deep learning-based image reconstruction pipeline. I made the training code easier to adapt to new scenarios, solved performance issues, 并制作评估报告供公司领导使用.
- 通过分析信号数据集和应用不同的数字信号处理(DSP)技术，改进了主动噪声消除(ANC)算法的性能. 因此，以前无法使用的数据集现在可以用于成像.
- 原型分析工具显著加快了MRI实验的分析时间.
Technologies: Python 3, PyTorch，数字信号处理，图像处理，Python, Computer Vision, Deep Learning
Senior Data Scientist and Machine Learning Engineer
2019 - 2022
Freelance
- 以端到端方式为推荐和其他机器学习系统做出了贡献. 与产品经理一起设计和创建早期降低风险的原型, built large-scale production systems, and designed A/B tests for evaluation.
- Coached team members, contributed to teamwork practices, 并提供了关于推荐和机器学习的实践研讨会.
- 就推荐系统的技术策略向产品经理和工程师提供建议.
Technologies: Spark, SQL, Google Cloud Platform (GCP)， Python 3, Machine Learning, Scala, Data Science, Machine Learning Operations (MLOps), Python
Senior Data Scientist and Machine Learning Engineer
2015 - 2019
SoundCloud
- Contributed to the core recommendations algorithms, including the matrix factorization, word2vec, factorization machines, locally sensitive hashing, learning to rank, and counterfactual evaluation.
- 使用气流对etl进行数据工程工作, SQL, Spark, Cassandra, APIs for online serving, and A/B testing infrastructure.
- 进行用户行为分析、实验设计评估等数据分析任务.
Technologies: Spark，机器学习，Python 3, Scala, SQL, C, c++， TensorFlow，数据科学, Machine Learning Operations (MLOps), Python, Computer Vision
Experience
多目标推荐:平衡消费者和内容提供者的利益
In this project, 将现有的推荐系统扩展到用户体验和内容成本之间的权衡.
我和一个产品组织负责人一起对这个项目进行了头脑风暴. We obtained positive evidence within one month. Right about three months into the project, 在生产中运行的坚实原型的A/B测试显示了积极的结果.
After that, 我与推荐团队合作，使这个原型可以投入生产, run a new A/B test, and confirm a positive impact on business.
我和一个产品组织负责人一起对这个项目进行了头脑风暴. We obtained positive evidence within one month. Right about three months into the project, 在生产中运行的坚实原型的A/B测试显示了积极的结果.
After that, 我与推荐团队合作，使这个原型可以投入生产, run a new A/B test, and confirm a positive impact on business.
Weekly Music Recommendation System for over 100 Million Users
I led a music recommendation project through conception, prototyping, implementation, and production deployment.
The project had a tight deadline of three months, a high-quality threshold, 大规模交付需求——每周交付超过1亿用户. 这个阈值对于将新内容与平台上其他现有推荐内容进行比较非常重要, and at the same time, the content had to be personalized.
The system was deployed to production. 用户和公司领导层对此反应非常积极. 许多平台用户声称，在这款推荐产品推出后，他们已经购买了订阅服务.
The project had a tight deadline of three months, a high-quality threshold, 大规模交付需求——每周交付超过1亿用户. 这个阈值对于将新内容与平台上其他现有推荐内容进行比较非常重要, and at the same time, the content had to be personalized.
The system was deployed to production. 用户和公司领导层对此反应非常积极. 许多平台用户声称，在这款推荐产品推出后，他们已经购买了订阅服务.
Audio-based Music Recommender
我指导一名实习生创建了一个基于音频数据的音乐推荐系统.
The motivation for this project was that content consumption was quite uneven on online platforms; most user consumption happened on a small proportion of items. As a result, 目录中的许多内容没有足够的信息来使用协同过滤技术.
我们建立了一个基于内容的推荐系统，从很少或没有消费信息的内容开始. 我们将此任务设置为从音频内容到现有协同过滤系统生成的推荐嵌入的回归任务. 为此，我们研究了音乐数据的频域表示. 输入数据是对音频信号进行mel变换后的短时傅里叶变换. 我们使用TensorFlow内置的卷积神经网络来生成推荐嵌入. 推荐检索是基于近似近邻索引，使用基于音频的嵌入.
The motivation for this project was that content consumption was quite uneven on online platforms; most user consumption happened on a small proportion of items. As a result, 目录中的许多内容没有足够的信息来使用协同过滤技术.
我们建立了一个基于内容的推荐系统，从很少或没有消费信息的内容开始. 我们将此任务设置为从音频内容到现有协同过滤系统生成的推荐嵌入的回归任务. 为此，我们研究了音乐数据的频域表示. 输入数据是对音频信号进行mel变换后的短时傅里叶变换. 我们使用TensorFlow内置的卷积神经网络来生成推荐嵌入. 推荐检索是基于近似近邻索引，使用基于音频的嵌入.
用于问答和领域特定信息检索的聊天机器人系统
用于问答和特定领域信息检索的聊天机器人系统
聊天机器人系统是为一个服务于少数群体需求的非政府组织开发的. 目标是帮助改善工作人员为非政府组织目标群体提供咨询的工作. It was developed in a very tight schedule of two months.
我和一位产品经理一起概述了需求, establish milestones, and develop the whole solution. 我们为聊天机器人使用了Rasa框架，并建立了一个易于维护的问答数据库. 该聊天机器人还与非政府组织的搜索引擎集成，以方便检索相关信息，并在规定的项目时间内成功部署.
聊天机器人系统是为一个服务于少数群体需求的非政府组织开发的. 目标是帮助改善工作人员为非政府组织目标群体提供咨询的工作. It was developed in a very tight schedule of two months.
我和一位产品经理一起概述了需求, establish milestones, and develop the whole solution. 我们为聊天机器人使用了Rasa框架，并建立了一个易于维护的问答数据库. 该聊天机器人还与非政府组织的搜索引擎集成，以方便检索相关信息，并在规定的项目时间内成功部署.
Skills
Languages
Python 3, OCaml, Python, Scala, C, SQL, C++
Frameworks
Spark
Libraries/APIs
TensorFlow, Rasa NLU, PyTorch, Scikit-learn, Pandas
Tools
Rasa.ai
Paradigms
Data Science, Functional Programming, Compiler Design
Other
Machine Learning, Machine Learning Operations (MLOps), Deep Learning, Computer Vision, Research, Data Analysis, Natural Language Processing (NLP), Digital Signal Processing, Image Processing, Benchmarking, Compilers, GPT, Generative Pre-trained Transformers (GPT)
Platforms
Linux, Google Cloud Platform (GCP)
Education
2006 - 2009
Ph.D. Degree in Computer Science
Utrecht University - Utrecht, The Netherlands
Collaboration That Works
How to Work with Toptal
在数小时内，而不是数周或数月，我们的网络将为您直接匹配全球行业专家.
1
Share your needs
在与Toptal领域专家的电话中讨论您的需求并细化您的范围.
2
Choose your talent
在24小时内获得专业匹配人才的简短列表，以进行审查，面试和选择.
3
Start your risk-free talent trial
与你选择的人才一起工作，试用最多两周. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring