Xiaoxi Li

Software Engineer

San Francisco, California, United States6 yrs 6 mos experience

Highly StableAI ML Practitioner

Key Highlights

Developed Alexa's Auto-Buy with 99% API accuracy.
Built real-time ML pipelines for fintech trading.
Expert in agentic AI and backend systems.

Stackforce AI infers this person is a Backend and Machine Learning Engineer specializing in Fintech and B2C AI solutions.

Contact

Skills

Core Skills

Machine LearningAgentic Ai DevelopmentSoftware Infrastructure

Other Skills

AWS LambdaKinesisDynamoDBReactMachine Learning AlgorithmsAWSXGBoostPythonKerasPandasNumpyMySQLRSoftware DesignInfrastructure

About

Software Engineer focused on backend + ML/LLM production systems, currently at Amazon Alexa Shopping building agentic AI experiences that turn natural language intents into real, reliable workflows. I developed Alexa’s agentic Auto-Buy end-to-end—LLM reasoning + deterministic routing + AWS serverless backend (Lambda/Kinesis/DynamoDB)—supporting 100K+ monthly interactions with sub-second latency and 99% API selection accuracy across multi-turn flows. Previously, I built real-time ML inference and streaming data pipelines in fintech, productionizing XGBoost valuation services on AWS with autoscaling, model versioning, monitoring, and sub-second market data ingestion to power trading decisions. I enjoy building systems that ship: high-throughput event pipelines, low-latency APIs, observability, and reliability improvements that reduce SEVs and improve customer experience. Open to backend/platform and ML infrastructure roles (SWE/MLE) where I can own core services from design to production.

Experience

Amazon

Software Development Engineer

Jan 2025 – Jan 2026 · 1 yr

At Amazon, I contribute to building production agentic AI and backend services that translate natural-language shopping requests into reliable workflows. I worked on Alexa’s agentic Auto-Buy experience, helping integrate the LLM reasoning layer with deterministic routing and downstream shopping services, supported by AWS components such as Lambda, Kinesis, and DynamoDB. I also contributed to improving system reliability and observability by restoring test coverage, addressing recurring operational issues, and building monitoring dashboards. In addition, I supported Shopping Essentials by delivering frontend improvements in React and collaborating on backend optimizations to reduce latency and improve state consistency across services. (LangChain, SG Lang)

Machine LearningAgentic AI DevelopmentAWS LambdaKinesisDynamoDBReact

Blue water financial technologies

Machine Learning Engineer

Nov 2022 – Dec 2024 · 2 yrs 1 mo

I built and productionized ML inference infrastructure and real-time data pipelines used for mortgage servicing rights (MSR) valuation and trading decisions. I deployed XGBoost-based valuation services on AWS with autoscaling, feature hydration, and monitoring, delivering low-latency APIs for internal users. I also integrated a Kinesis streaming pipeline to ingest live market signals with sub-second updates, reducing stale-price risk and improving decision speed. In addition, I developed core MLOps components for model versioning, drift monitoring, and retraining triggers to support a scalable, production-grade ML platform.

Software InfrastructureMachine Learning AlgorithmsAWSXGBoostKinesisMachine Learning

University of minnesota

2 roles

Research Assistant

Promoted

Sep 2019 – Jul 2022 · 2 yrs 10 mos

The Comparative Data Analysis of Trading Regulations (Python, Keras)
Acquired trading data between China and the U.S. and cleaned it to a structured format using Python (Pandas, Numpy), preprocessed the data to remove invalid entries, and matched them with corresponding labels.
Predicted international agricultural products trade volume based on data from 2000 to 2018 with a pretrained CNN, achieving an accuracy of 85% in most genres after parameter tuning.
Analyzed the relationship among international trade, foodborne illnesses, and food safety governance system by variable correlation and importance analysis, reached a conclusion that reveals the dominant factors that result in the frequency of food safety issues.

Research Assistant

Sep 2017 – May 2018 · 8 mos

Wheat Functionality Prediction (MySQL, R)
Efficiently cleaned data sources and constructed MySQL database summarizing the indicators of wheat functionalities with different fertilization rates of 3 locations in Minnesota.
Clustered the functionality data with the K-Means algorithm and the optimal number of clusters is determined using the elbow approach.
Visualized the clusters by reducing the original data to lower dimensions using PCA and t-SNE, revealing the intrinsic difference after applying multiple fertilization conditions
Published paper: Effect of sulfur fertilization rates on wheat (Triticum aestivum L.) functionality.