HELLO, I’M

Caroline Kim

I build data and machine-learning systems end to end — scraping and cleaning real-world data, modeling it, and shipping the result as tested, containerized, cloud-deployed services.

Over one summer at Johns Hopkins (EP 605.256, Modern Software Concepts in Python) I took a single messy dataset — 100,000 graduate-admissions records — from raw scrape to a fine-tuned DistilBERT model deployed behind a Flask app, hardening every layer along the way: tests at 100% coverage, injection-safe SQL, Docker microservices, AWS deployment, and tracked ML experiments.

See the work → Get in touch

CORE SKILLS

The tools I reach for most, and what I’ve built with them.

Cloud & DevOps

Docker Compose Four-service stack decoupled by a RabbitMQ queue and background worker AWS S3 to SageMaker data pipeline, plus the full stack deployed on EC2 pytest 100% coverage, marker-organized, run in CI against a real database

Data Engineering

PostgreSQL Loaded ~100k scraped records with idempotent, concurrency-guarded writes pandas Cleaned 99,941 rows to 96,948 with an auditable ledger of every drop NumPy Implemented a two-layer neural network and its gradients from scratch

ML & Modeling

PyTorch Fine-tuned DistilBERT for admissions prediction and deployed it live scikit-learn TF-IDF, PCA and K-Means over 5,861 messy program names MLflow Tracked parameters, metrics and fitted models across training runs

Data sources ClinicalTrials.gov NCI Thesaurus

FOCUS

Cloud & DevOps

Testing & CI, secure SQL, Docker microservices, AWS (S3, SageMaker, EC2).

Data Engineering

Scraping, cleaning with audit ledgers, PostgreSQL, idempotent pipelines.

ML & Modeling

Clustering, neural networks from scratch, LM fine-tuning, experiment tracking.