Ruike (Oliver) Hou
Knowledge Graph
Interactive visualization of my skills, projects, and experiences. Filter the map, inspect a node, then ask the AI why it matters.
My Journey
Bachelor of Science in Computer Science
University of Toronto
"CS Specialist Co-op student with a Statistics minor, focused on AI/ML, HCI, software engineering, CV, and data visualization."
Program
CS Specialist Co-op
Minor
Statistics
AI Engineer Intern
Remeda
"Building healthcare AI systems across RLHF, multimodal respiratory anomaly detection, and clinical backend standards."
Interpretability
+15%
False positives
-22%
Clinical audio
5,000+ hrs
F1 score
89%
Simulated cases
700+
Clinical AI systems need reasoning that is easier to audit while reducing noisy alerts that slow down care teams.
Built an RLHF workflow for diagnostic reasoning, fused audio and text signals with PyTorch, and mapped backend documentation to FHIR and SNOMED-CT standards.
Improved interpretability scores by 15%, reduced false-positive alerts by 22%, reached 89% F1 on early respiratory anomaly detection, and projected 1.5 hours of daily paperwork savings for doctors.
Apple Intelligence Glow Effect
Open Source NPM Package
"Published npm package recreating Apple Intelligence's animated glow border effect."
Package
npm
Format
React component
Developers often want polished AI-era visual effects without rebuilding complex gradient animation systems from scratch.
Packaged layered conic-gradient glow logic into reusable React components with customizable containers and demo surfaces.
Turned a visual interaction pattern into an open-source package that can be reused across portfolio and product interfaces.
Aurora Loyalty Web Application
CSC309 Course Project
"Full-stack loyalty program with AI assistant and role-based access control."
Stack
React + Express
AI
Gemini assistant
A loyalty platform needs different workflows for customers, organizers, and admins while still feeling coherent to users.
Built role-based access, points redemption, RSVP flows, recommendations, dispute handling, and an embedded Gemini assistant.
Delivered a complete full-stack application with practical AI support layered into normal user workflows.
LLM Algorithm / Full Stack Intern
Nowcoder Co., Ltd
"Optimized AI interview-answer detection, trained fallback classifiers, and built a conversational AI+BI platform."
LLM detection
90.5% acc
BERT fallback
88% acc
BERT recall
85%
Screen classifier
98.7% acc
Screen recall
91.3%
Interview platforms need to detect AI-assisted answer completion while keeping fallback detection reliable when LLM prompts are insufficient.
Combined prompt tuning, Claude Sonnet 4 evaluation, a BERT fallback classifier, and an EfficientNet-B3 screen-content model; then connected AI reasoning to BI workflows with text-to-SQL and chart generation.
Reached 90.5% LLM detection accuracy, 98.7% screen classification accuracy, and shipped a conversational Java21/Vue3 BI platform that executes SQL and produces ECharts visualizations.
PhysioApp – AI-Integrated Fitness App
Course Project Leader
"Full-stack fitness platform with DeepSeek R1 API integration."
Stack
React + Spring
AI API
DeepSeek R1
Ops
Docker + CI/CD
Fitness apps often ignore injury context, so users may receive generic plans that are not suitable for rehabilitation.
Led a full-stack build with injury-aware onboarding, AI guidance through DeepSeek R1, dynamic muscle visualization, and workout/rehab tracking.
Created a personalized fitness workflow that connects recommendation, explanation, and progress monitoring in one application.
MB-FIRE: Dual-Channel Reconstruction
Researcher with Prof. Rabih Younes, Duke University
"Dual-channel frequency-guided reconstruction for AI-generated image detection."
AUROC lift
+18.1
Classifier
9-channel ResNet-50
Datasets
SDXL/MJ/K3
AI-generated image detectors often overfit to one generator and fail to generalize across newer diffusion models.
Combined mid-band frequency reconstruction cues with high-pass spatial residuals, then classified concatenated error maps with a 9-channel ResNet-50.
Improved macro-average AUROC by 18.1 over SOTA baselines such as FIRE, with stronger cross-model generalization on SDXL, Midjourney, and Kandinsky 3.
Database ERP Intern
Infosys Technology Co., Ltd
"Supported Oracle ERP deployment through data-flow analysis and migration preparation."
Records
50,000+
Business units
5+
ERP migrations are fragile when source records are inconsistent across business units.
Studied Oracle ERP module flows and prepared migration data through cleaning, validation, and go-live support.
Helped prepare 50,000+ records for a deployment affecting 5+ business units.
Software Engineering Intern
Shanghai Wangmeng (Alibaba)
"Maintained Spring Boot microservices and deployed a LoRA-tuned Mistral-7B emotion recognition workflow."
Emotion model
94.3% acc
Model
Mistral-7B
Customer-service systems need emotion-aware responses while staying maintainable inside microservice infrastructure.
Maintained Spring Boot services with Maven, optimized AWS usage, and deployed a locally hosted LoRA-tuned Mistral-7B model.
Delivered emotion recognition and response functionality with 94.3% accuracy.
Graphical Analysis of NBA Behaviors
Academic Project
"Graph theory analysis of collaborative behaviors in basketball."
Method
Graph theory
Library
NetworkX
Basketball collaboration patterns are difficult to inspect from box scores alone.
Modeled player interactions as a graph and used NetworkX visualizations to reveal strategic relationships.
Produced a clearer analytical view of team behavior and collaboration patterns.
SDSS Datathon: Airfare Markets Under Pressure
2026 SDSS Datathon Participant
"Built analytics-dashboard work for airfare market pressure analysis."
Event
SDSS 2026
Focus
Airfare analytics
Airfare markets shift under pressure from demand, route constraints, and pricing changes.
Worked on dashboard-oriented analysis to surface market patterns and support fast comparison.
Added applied analytics and visualization experience beyond coursework and internships.