Hi, I'm Songyan Zhao

Machine Learning Engineer & Full-Stack Developer

4+ years of ML/AI experience. Skilled in LLMs, agentic AI, RAG systems, recommendation systems, human-AI interaction, and training/finetuning models. First-author publications at top NLP conferences (NAACL, EMNLP, etc.).

7+ years of full-stack experience skilled in Java (Spring Boot) and Python (Flask) for scalable distributed systems, RESTful APIs, and AWS services, with frontend expertise in React/React Native, Vue.js, and TypeScript.

Currently seeking full-time Machine Learning Engineer or Software Engineer roles starting November 2025. If there is a good fit, please reach out!

MLE Resume Full-Stack Resume Get In Touch
Songyan Zhao

Work Experience

Building cutting-edge AI systems and research

Machine Learning Engineer Intern

XPOWER Manufacture Inc. | AI Infrastructure - Vendor-Recommendation Agent

July 2025 - Present
  • Built a unified vendor-recommendation platform with React frontend and Flask API gateway orchestrating Amazon Bedrock Agents, Bedrock Flows, and AWS Lambda for intelligent service provider discovery
  • Built dual-path search architecture—internal RAG via Bedrock Knowledge Bases for certified partners and external discovery via SerpAPI (Google Maps/Yelp)—supporting parallel execution, automatic failover, and 504 timeout retries
  • Automated infrastructure deployment including Bedrock Flow provisioning, IAM role configuration, and cross-account Lambda permissions, enabling secure execution of 1, search; 2, enrichment; and 3, ranking pipeline
  • Enhanced search accuracy with LLM-based parameter extraction (Claude Sonnet 4) that converts natural language requests into structured Lambda inputs, complemented by trace logging to debug agent action-group invocations

Software Developer Intern

Xpower Inc. | Spring Boot Based ChatRoom with AI-assistant

Apr 2025 - Aug 2025
  • Built a Spring Boot chat backend (WebSocket/STOMP, Redis/Postgres) with Twilio fallbacks, S3 uploads, and AI-assistant JSON card messages
  • Designed the Project → Item → Room domain model and implemented role-scoped REST APIs with Redis-backed delay/archival jobs to maintain unread counts, SLAs, and AI-triggered workflows
  • Delivered an Expo/React Native client with per-role project/item/chat flows and inline AI-assistant activity
  • Improved mobile reliability via typed OpenAPI clients, Zustand chat state, and a resilient SockJS/STOMP bridge

Research Assistant

UCLA PlusLab & Queen Mary University of London

Sept 2023 - Present
  • Led the design of REFFLY, the first melody-constrained lyric revision model based on Llama 2, improving fluency and musicality by 25% over strong baselines. First-author paper accepted at NAACL 2025 (oral presentation)
  • Co-developed VDebugger, a critic-refiner system for visual program debugging. Improved error localization and correction accuracy by 3.2% across six visual reasoning benchmarks. Published at EMNLP Findings 2024
  • Built an AI-assisted lyric writing interface based on REFFLY, with FastAPI (backend) and Vue.js + Vite (frontend); paper under submission to UIST 2025
  • Built a Variational Autoencoder (VAE) using PyTorch for symbolic music generation with interpretable latent control, published in Machine Intelligence Research (IF: 6.4)

Software Developer Intern

Baidu International Department (Do-global) | Spring Boot E-commerce Mobile APP

June 2021 - Sept 2021
  • Implemented REST API via Spring MVC including user registration/login, product listing and search, cart management, order placement, payment processing, and user profile updates
  • Utilized Spring Data JDBC for PostgreSQL database integration, managing products, users, carts, and orders
  • Implemented Spring Security for session-based authentication and authorization
  • Used the Spring framework core technologies to loosely decouple all the components in the application
  • Developed the client side using ReactJS and Ant Design, enabling users to seamlessly add items to their shopping carts and place orders

Technical Skills

Technologies and tools I work with

Languages

Python Java C C++/STL SQL JavaScript TypeScript Go HTML/CSS

Frameworks & Tools

LangChain PyTorch Scikit-Learn AWS Lambda AWS Bedrock OpenAI API Claude API Spring Boot Flask React FastAPI Vue.js Node.js

Databases & Tools

MySQL PostgreSQL MongoDB Redis Core Data Docker Git Maven Gradle Dplr gdb

AI & ML Expertise

Large Language Models Agentic AI RAG Systems Training/Finetuning NLP Machine Learning Deep Learning Recommendation Systems Human-AI Interaction

Research Publications

AI/ML research with first-author publications at top venues

REFFLY - AI Lyric Generation

Led the design of the first melody-constrained lyric revision model based on Llama 2, improving fluency and musicality by 25% over strong baselines. First-author paper accepted at NAACL 2025 (oral presentation). Built an AI-assisted lyric writing interface with FastAPI (backend) and Vue.js + Vite (frontend); paper under submission to UIST 2025, a top-tier HCI venue.

Llama 2 NLP FastAPI Vue.js PyTorch NAACL 2025
Read Paper

VDebugger - Visual Program Debugging

Co-developed VDebugger, a critic-refiner system for visual program debugging. Improved error localization and correction accuracy by 3.2% across six visual reasoning benchmarks. Published at EMNLP Findings 2024.

LLMs Visual Reasoning Python NLP EMNLP 2024
Read Paper

Symbolic Music Generation with VAE

Built a Variational Autoencoder (VAE) using PyTorch for symbolic music generation with interpretable latent control by regularizing latent vectors. Published in Machine Intelligence Research (Impact Factor: 6.4).

VAE PyTorch Music Generation Deep Learning Machine Intelligence Research 2024
Read Paper

Featured Projects

Software engineering and development projects

AI-analyst: LangChain RAG Retrieval & QA System

Created an interactive conversational UI leveraging React and Ant Design, enabling users to upload and interact with PDF, Excel documents in real-time. Implemented RESTful APIs via Express and Node.js and optimized for high-performance request handling. Utilized an in-memory vector store to cache generated embeddings for efficient retrieval. Integrated OpenAI's GPT-4o API and Langchain to develop an advanced AI agent for document analysis, loading, splitting, storage, retrieval, and output.

React Node.js Express GPT-4o LangChain RAG Vector Store Ant Design

MiniSpotify

Designed a Kotlin Spotify favorite app by fully leveraging Android Jetpack Library and Hilt Dependency Injection. Implemented the BottomBar and app navigation using the Jetpack Navigation component. Created a mock RESTful API with json-server and handled requests using Retrofit. Built the feed/album/favorite UI in Jetpack Compose following MVVM architecture. Enabled local cache for favorite features with Room Database, and integrated Google ExoPlayer for global music playback.

Kotlin Jetpack MVVM Room DB Retrofit ExoPlayer Hilt

SocialAI

Designed and implemented an AI-based social network web application using React JS. Integrated OpenAI's DALL-E 3 to assist users in creating and updating posts. Improved authentication with token-based registration, login, and logout via React Router v4 and server-side JWT. Developed and deployed a scalable web service in Go for post management on Google Cloud (Google App Engine). Deployed ElasticSearch on GCE to enable search for recent and personal posts.

React Go DALL-E 3 JWT GCP ElasticSearch

Education

Academic background

University of California, Los Angeles

M.S. Computer Science

Sept 2023 - June 2025 GPA: 3.88/4.0 Advisor: Prof. Nanyun Peng

Carleton College

B.A. Computer Science and Mathematics (Double-majors)

Sept 2019 - June 2023 GPA: 3.86/4.0, Cum laude

Get In Touch

Email Me Connect on LinkedIn
zhaosongyan7@gmail.com
+1-507-581-1895
Los Angeles, CA