Annie Yu

Hi! I build friendly AI — agents, knowledge graphs & machine-learning systems.

scroll

3 terms at IBM CS @ Western · HBA @ Ivey

01 About

I build AI systems that remember — multi-agent pipelines, knowledge graphs, and the models in between. I study Computer Science at Western and business at Ivey, and I love building tools that people actually use.

3.8
GPA · Western CS
3
Terms at IBM
70%
Faster error diagnosis with GraphRAG agents

Education

Western University
HBSc Computer Science · GPA 3.8
2022 — 2028

Ivey Business School
HBA, combined degree
2025 — 2027

Leadership

VP of Projects, Western Founders Network

Co-Captain, Ultimate intramural team

Ivey Tech Club · Ivey Product Society

Recognition

Western Scholarship of Distinction

Dean’s Honor List

North America Cup winner — Team Canada, Golf

02 Experience

May — Aug 2026

IBM Term 3

Application Developer Intern

  • Built a multi-agent UI-to-code pipeline on Microsoft Agent Framework with persistent Cognee/Neo4j memory, raising component-match accuracy by 40%.
  • Designed and presented a graph-database POC to a client, cutting manual lookup time by 60%.
  • Benchmarked LLM & embedding providers and designed a 7-stage schema to scope agent memory retrieval.

May — Aug 2025

IBM Term 2

Application Developer Intern

  • Developed agentic tools combining Neo4j GraphRAG and Claude 3.7 Sonnet for error diagnosis, cutting developer response times by 70%.
  • Evaluated RAG architectures & agent frameworks; built Storybook UIs for agent reasoning.
  • Architected a Crawl4AI/Puppeteer pipeline extracting 10K+ material specs for a shipbuilding client.

Sep — Dec 2024

IBM Term 1

Developer Intern

  • Engineered validation rules for an anomaly-detection engine verifying $10B in insurance claims on Azure Synapse.
  • Ran parallel QA on two validation engines, resolving discrepancies across 50M+ rows.

Feb 2023 — Jul 2024

TMU

ML Research Assistant · Dept. of Mathematics

  • Developed RL market-making agents (DDPG, DQN) that learn Bitcoin bid/ask quoting strategies.
  • Built a custom trading environment on 25-level limit order book data; actor-critic networks in PyTorch & TensorFlow.

03 Selected projects

Things I’ve built

Computer vision
92%

crowd counting accuracy · 15 FPS

Crowd Watch

Real-time crowd density, from the sky.

CSRNet with a VGG16 backbone and dilated convolutions on DJI drone & webcam feeds — density heatmaps for 10K+ person crowds, alerting above 5 people/m².

  • PyTorch
  • CSRNet
  • OpenCV
  • DJI Drone
Wearable · LLM
Live

scene mood → playlist

Mookie

A wearable that picks your soundtrack.

A Raspberry Pi Zero 2 camera streams snapshots to Flask, where a locally hosted Llama 3.2 Vision model reads the mood of the scene; a LangChain pipeline curates live Spotify playlists.

  • LangChain
  • Ollama
  • Llama 3.2 Vision
  • Spotify API
IoT · Full-stack
500+

students during exams

Room Finder

Find an empty study room, instantly.

Arduino sensors across 30+ rooms with 95% detection accuracy, occupancy probabilities computed from 10K+ records, and a PostgreSQL + REST backend feeding live React visualizations.

  • React
  • PostgreSQL
  • Arduino
  • REST
Archive
More

on GitHub

@annooee ↗
PythonPyTorchNeo4j GraphRAGLangChainAzure OpenAIReactSQL
Agent FrameworkCogneeTensorFlowOllamaAWS BedrockDocker

Languages

Python, Java, SQL, JavaScript, C, R, HTML/CSS

Frameworks

React, Node.js, Flask, PyTorch, TensorFlow, Stable Baselines3, LangChain, Pandas, NumPy, OpenCV

AI & Data

Microsoft Agent Framework, Cognee, Neo4j (GraphRAG), AWS Bedrock, Azure OpenAI, Ollama, Azure Synapse

Tools

PostgreSQL, Docker, Storybook, Power BI, Puppeteer, Crawl4AI, Raspberry Pi, Arduino