Yuji Oshima

•Designed and developed a robust and scalable MLOps pipeline from data ingestion to model deployment using Kubeflow, MLflow, and Seldon Core on EKS. •Improved the security of the system by following best practices such as RBAC(Role-Based Access Control), integrating a service mesh like Istio, and applying network policies on the Kubernetes cluster. •Automated the ETL data pipeline using Apache Airflow, Kafka, Spark, and Tableau, enhancing scalability through best practices. •Developed a RAG chatbot utilizing LangChain and deployed Mixtral Large 7 B on SageMaker Jumpstart using Hugging Face LLM DLC, powered by TGI on AWS. •Created a hotel recommendation system using the HyDE approach with Redis and the OpenAI API. •Fine-tuned the LLM for a customer support chatbot using QLoRa and ORPO techniques on Amazon SageMaker Jumpstart, implementing canary deployment. •Integrated a Large Language Model (LLM) into resume matching and analysis using LangChain and executed Llama 7B on Azure via Azure ML. •Designed a wide range of prompt techniques, such as chain-of-thought and step-back prompting, to improve model performance.

Developed and maintained a vector-search-based recommendation system on AWS using Elasticsearch for an e-commerce platform, resulting in an 18% increase in customer engagement and a 12% increase in sales. •Built the MLOps pipeline for the recommendation system with Amazon SageMaker Pipeline on AWS and integrated MLflow to track the model's performance. •Fine-tuned CLIP (Contrastive Language-Image Pre-Training) to significantly boost the performance of item category and brand prediction. •Implemented a CI/CD pipeline for training and prediction using Amazon SageMaker for the detection of chargeback transactions in the Merpay payment service, conducting cancel detection of transactions based on chat messages with NLP. •Enhanced MLOps pipelines for the deferred payment service with manual confirmation and implemented model management for client-side ML powered by TensorRT and Firebase. •Conducted neural architecture search (NAS) to automate multi-modal modeling for prohibited item detection. •Used Tableau and SQL to redefine and track KPIs surrounding marketing initiatives and implemented A/B tests to generate 15% more client leads. •Reduced fraudulent activities by 49% by creating an anomaly detection system for financial transactions to safeguard the companys assets. •Collaborated with cross-functional teams to gather and preprocess data, fine-tune models, deploy several models into production environments, and power client-side machine learning with TensorFlow Lite.

Developed a serverless architecture using AWS Lambda and API Gateway to achieve a cost-effective and scalable backend solution. •Managed AWS infrastructure, including EC2 instances, S3 buckets, and RDS databases, ensuring high availability and data security. •Built and maintained back-end server-side applications using Node.js, ensuring efficient communication between the Server and Client-side components. •Enhanced an e-commerce site's search capabilities by leveraging Algolia to provide users with fast and relevant search results. •Implemented user authentication and authorization using industry-standard protocols like OAuth2 and Clerk, resulting in improved security and reduced a risk of data breaches. •Developed a RESTful API using Node.js and Express, enabling the company to expose its services to external users and systems, resulting in improved efficiency and flexibility. •Leveraged Salesforce CRM to streamline customer relationship management, enhancing sales and marketing processes. •Implemented Semantic HTML5 markup throughout the project, enhancing accessibility and improving Search Engine Optimization ( SEO ) by providing clear and meaningful structure to web content.