About Me
Hi, I'm Erica Zhao — a Machine Learning Engineer focused on LLMs evaluation and applied AI for real user workflows. I build end-to-end ML/AI solutions from data pipelines to model iteration, and I love turning ambiguous problems into reliable, measurable systems.
Recently, I've been building LLM-powered features for education workflows—designing prompts and rubrics, enforcing structured outputs, and running evaluations to improve quality, consistency, and user trust. I'm especially interested in applied GenAI where product goals and measurable model behavior meet.
Let's connect if you'd like to collaborate, discuss technology, or just say hello!
Projects
Adaptive AI Assistant
Built and iterated LLM-powered workflows for leaning assistant: prompt + rubric design, structured outputs, evaluation criteria, and quality analysis against real user tasks. Focused on reliability, consistency, and measurable improvements. (Company project; details are confidential.)
Mental Health Demand-Supply Gap Monitor
An interactive data-to-dashboard system that tracks mental health demand signals vs provider capacity across U.S. states (2020-2024). Interpretable ML highlights where gaps are largest and what factors drive them.
Wallet Wizard - Personal Finance Tracker
Full-stack personal finance web app with authentication, CRUD budgets/transactions, and a REST API. Built a responsive dashboard in Next.js/React, backed by Node.js services and PostgreSQL for reliable expense tracking and monthly budgeting.
NIH Grant Activity Classifier
Trained a multi-class model to predict NIH grant activity series (F/T/K/R) from structured award signals using public NIH RePORTER data (FY2018–FY2023, 118k+ records). Best baseline achieved ~0.93 macro-F1 on validation.
Test Data Refresh Pipeline
Built a data pipeline to consolidate and analyze security-related datasets, enabling faster monitoring and investigation. Emphasized data quality, reproducibility, and scalable processing for downstream analytics.
EdTech Support Chatbot
Built a modular EdTech support chatbot (easy → advanced): NLTK rules, intent classification, and an LLM prototype. Compared reliability/latency/maintainability tradeoffs and documented an upgrade path from GPT-2-era prototypes to modern LLMs with guardrails.
Options VaR Risk Engine
A reproducible Python risk engine that uses a Gaussian copula with Student-t marginals to simulate joint SPX-VIX scenarios, re-price an options butterfly spread via Black-Scholes, and compute 1-day 95% VaR from the simulated PnL distribution.
Decentralized Lottery System Smart Contract
A decentralized lottery system implemented as a Solidity smart contract on the Ethereum blockchain. The contract allows users to enter the lottery by sending Ether, and a random winner is selected at regular intervals using Chainlink VRF for verifiable randomness. The project demonstrates secure and transparent lottery mechanics while leveraging blockchain technology.
Tech Stack
Global Education Journey
A snapshot of where I studied, what I focused on, and the milestones that brought me here.
- Machine Learning
- Artificial Intelligence
- Software Development
- Machine Learning
- Algorithms Design
- Data Visualization
- Foreign Direct Investment
- Corporate Strategy
- Systems Analysis & Design
- Database Management
- International Marketing
- Consumer Behavior
Writing
Short notes + longer essays from building and events.
Events & Volunteering
Photos + short notes from tech events and volunteering.
IMF/ World Bank Spring Meetings — Volunteer
Supported sessions and logistics; observed discussions around AI, finance, and regulation.
Microsoft AI Tour — San Francisco
Focused on applied GenAI + ML systems; met builders and explored trust & safety themes.