Hoyeon Luke Jang

Incoming Data Engineer Intern

?? @ TD Bank

Got extended!

CollectLite

Personal Project

A lower-scale human-data collection and model-evaluation platform inspired by public evidence about Cohere's Collect / Human Data tooling. The system lets project owners upload datasets, create annotation tasks, use model-in-the-loop suggestions, collect human feedback, resolve disagreement, monitor data quality, and export training/evaluation-ready datasets.

• full-stack
• internal-tools
• human-data
• annotation-platform
• model-evaluation
• llmops
• nextjs
• typescript
• react
• fastapi
• python
• postgres
• cohere
• docker

Deep Concept Model (DCM)

Independent Research Proposal

A proposed neuro-symbolic AI research programme for building a formal concept algebra over mathematics, using dependent type theory, Lean 4, Mathlib, ontology graphs, hyperbolic embeddings, and verified concept generation to explore reasoning beyond natural language.

• research
• neuro-symbolic-ai
• knowledge-representation
• ontology
• type-theory
• lean4
• mathlib
• theorem-proving
• concept-generation
• ai-for-mathematics
• compositional-reasoning

Data Engineer Intern

AI Platform @ TD Bank

Maintained and enhanced production AI ETL, RAG ingestion, Azure infrastructure, model deployment, and observability systems for enterprise document intelligence workflows.

• azure
• ai-etl
• azure-sql
• azure-databricks
• azure-machine-learning
• azure-openai
• datadog
• ci-cd
• llmops

Predicting Online Shoppers’ Purchasing Intention

Western University

An end-to-end supervised machine learning pipeline for predicting purchase intent across 12.3K e-commerce sessions using behavioral feature engineering, leakage-safe preprocessing, class-imbalance-aware evaluation, and tuned linear, tree-based, and ensemble classifiers.

• machine-learning
• data-science
• classification
• ecommerce
• feature-engineering
• scikit-learn
• xgboost
• random-forest
• logistic-regression
• model-evaluation

Video Virality Predictor

Western AI

A multimodal YouTube Shorts forecasting platform that collects short-form video data, generates video/audio/text embeddings, fuses modalities, trains leakage-resistant predictors, and serves 7-day and 30-day view forecasts through a conference-demoed web application.

• machine-learning
• multimodal-ai
• forecasting
• youtube-shorts
• aws-s3
• pytorch
• scikit-learn
• nlp
• computer-vision
• audio

Building a GPT-Style LLM from Scratch

Personal Project

A ground-up PyTorch implementation journey for understanding GPT-style language models, covering tensor operations, autograd, neural network training loops, tokenization, byte-pair encoding, embedding construction, positional embeddings, causal self-attention, and multi-head attention.

• machine-learning
• deep-learning
• llm
• transformer
• gpt
• pytorch
• tokenization
• embeddings
• attention
• neural-networks
• nlp

Applied AI Developer Intern

Growth R&D @ Archisketch

Developed and deployed production-ready AI agents, recommendation systems, retrieval pipelines, and data infrastructure for Archisketch’s applied AI product features.

• python
• applied-ai
• ai-agents
• rag
• vector-databases
• graph-databases
• recommendation-systems
• prompt-engineering
• llamaindex
• neo4j
• chroma
• pinecone