Software Engineer focused on AI applications, developer tools, and full-stack systems. Recently shipped production features at startups and currently building a Personal Email Agent to explore AI-powered personal productivity systems.

San Jose, CA · B.S. Mathematics & Computer Science, UC San Diego

Personal Email Agent — A full-stack system for managing and understanding large inboxes (22,000+ emails). Built to turn email overload into structured, actionable context.

Architecture: Next.js frontend → FastAPI backend → service layer (Gmail sync, classification, analysis) → SQLAlchemy/SQLite persistence.

Current stack: Python, FastAPI, Pydantic, SQLAlchemy, Next.js, TypeScript, Tailwind CSS, Google Gmail API, OpenAI SDK.

Engineering decisions:

• Layered architecture with stable API contracts so mock data, rule-based classifiers, and future LLM analysis can swap without frontend rewrites
• Gmail OAuth + read-only sync service separated from analysis logic
• Typed schemas between frontend and backend for email, analysis, and Gmail payloads

Direction: Evolve from inbox analysis into a personal operating system — priority detection, summarization, reply drafts, follow-up tracking, and agent workflows with humans in the loop.

Sep 2025 – Dec 2025 · Silicon Valley, CA

• Shipped production features on the startup web platform in a fast-paced environment, working directly with the tech lead on new functionality and data persistence
• Owned end-to-end development of HerCommunity — forum discussions, events, and stories with full user and admin workflows (CRUD, auth, role-based access, data integrity)
• Debugged TypeScript backend services and REST APIs against PostgreSQL; improved reliability through manual regression testing across feature iterations

Sep 2024 – Jun 2025 · San Diego, CA

• Optimized deep learning models (EfficientNet, Fast-SCNN, MobileUNet, UNet) for real-time quantitative ultrasound image segmentation using PyTorch, TensorFlow, and Keras
• Built preprocessing pipelines for multi-format ultrasound datasets — normalization, augmentation, resizing, and sampling — to improve training efficiency and model performance
• Integrated PySide6 GUI, Blender-based 3D probe visualization, and Clarius Cast API for live ultrasound capture and real-time analysis workflows

Jan 2025 – Mar 2025 · San Francisco, CA

• Built a full-stack AI storytelling web app with Streamlit, LangChain, and OpenAI GPT — users generate multi-page personalized children's books with dynamic characters and adaptive illustrations
• Designed prompt pipelines with RunnableSequences and DALL·E 3; optimized backend architecture and API performance to improve text-to-image coherence and cut response latency by 28%
• Delivered the product from early MVP to deployment, managing technical documentation and phased releases for a production-ready LLM application

Problem: Large inboxes are hard to triage — important messages get buried and context is scattered across threads.

Stack: FastAPI, SQLAlchemy, Next.js, TypeScript, Gmail API, OpenAI SDK

Decisions: Service-layer separation for Gmail ingestion vs. classification vs. analysis; Pydantic schemas as the contract boundary; modular routers so AI analysis can replace rule-based classifiers incrementally.

Problem: Parents and educators need personalized, illustrated children's books without manual layout or illustration work.

Stack: Streamlit, LangChain, OpenAI GPT, DALL·E 3, RunnableSequences, Requests, PIL

Decisions: Chained LLM + image generation pipeline with prompt engineering for character consistency across pages; backend optimizations that reduced end-to-end latency by 28% while maintaining illustration coherence.

Problem: Course-specific questions are poorly served by generic chatbots that lack access to lecture materials, assignments, and syllabus context.

Stack: Python, LLM APIs, RAG over course documents, web UI

Decisions: Retrieval-grounded responses scoped to instructor-provided materials; structured prompt design to keep answers tied to course content rather than open-ended generation.

Problem: Text-to-image workflows need controlled prompts, batch generation, and post-processing — not just a single API call.

Stack: Python, OpenAI DALL·E, PIL, REST APIs

Decisions: Prompt template system for consistent visual style; image post-processing pipeline for sizing and format normalization; async request handling to manage API rate limits and latency.

Languages Python · TypeScript · JavaScript · Java · C++ · SQL · Bash

Backend FastAPI · Node.js · REST APIs · PostgreSQL · SQLAlchemy · SQLite

Frontend React · Next.js · Tailwind CSS · Streamlit

AI / LLM PyTorch · TensorFlow · Keras · LangChain · OpenAI API · DALL·E · RAG pipelines

Data NumPy · scikit-learn · data preprocessing · feature engineering · model evaluation

Cloud & DevOps AWS · Docker · Vercel · GitHub Actions · CI/CD · Linux

• LinkedIn: linkedin.com/in/christine-wu-2bab27243
• Email: cwu20210923@gmail.com