Lowan Kim
resume.pdf

I ship research ideas as a web tool

Senior engineer

Eight years of Python and Django, shipping tools licensed to pharmaceutical partners.

--year --tag

2026

2025

ship

Allele name search

slow searches → 98.5% faster

Search over 92,000 allele names, with a custom ranking and the data reshaped to serve it. Load time down 98.5%, and a Swagger-documented API so other tools and pipelines can call it.

live
ship

CLI tool framework

twelve setups → one framework

The Python framework every in-house CLI tool is built on, current and future: argument parsing, scaffolding, environments, dependencies and packaging, automated. Setup and onboarding time down 80%. Five tools migrated, seven under way.

in progress

2024

ship

BCRMatch

one notebook model → CLI and web app

An antibody-similarity model, built by a computational biologist, packaged as a CLI and a web app the whole lab can run. Reviewing the data turned up duplicate and misaligned records; she corrected the model before release.

internal

2023

ship

Legacy bridge

three tools stranded → three containers

Three legacy Python tools that could not run on modern servers, containerized one by one, with a Django app brokering requests through Gunicorn and the Docker SDK. Nobody lost access during the server updates.

live

2022

ship

Containerization

legacy VMs → docker-compose, 50% smaller

The 40 GB IEDB Tools site, fifty-odd interdependent apps, moved from VMs to Docker and docker-compose. Deployment footprint down 50%, and the institute’s first standard DevOps workflow.

live

2021

ship

InputManager

per-tool checks → one module, 21 schemas

A reusable Python module that validates and handles input for 21+ schemas, so each tool stops carrying its own copy of the same code.

internal

2019

ship

Licensed releases

one server, ours → OVF or Docker, theirs

Twenty-two official releases of IEDB tools licensed to AstraZeneca, Vertex, Amgen, Moderna and Pfizer: triage, development, packaging, deployment and support, as standardized VMware OVF and Docker builds. Four hundred-odd support tickets along the way.

shipped

2017

ship

Sequence models

two feature sets → one model, 95%

Machine-learning models that predict patterns in sequence data, at the San Diego Supercomputer Center. Merging complementary features with two other engineers took accuracy to 95% and F1 to 75% on the elements that mattered.

research

lowan --all