HELLO, I’M
Software Developer
I build reliable backend systems for payments and operations, automate repetitive work, and connect Python·FastAPI with AI to improve real workflows.
01ABOUT ME
From a public library classroom to building AI products. These are the places that shaped how I write software, and what each one taught me.
SCROLL TO FOLLOW THE JOURNEY
Through the New York Public Library's TechConnect program, I learned programming fundamentals in a classroom of people from every background. It was the first time I built something that actually ran.
WHAT I LEARNED
A Chingu Voyage put me on a remote team with developers I had never met, a product to ship, and a fixed number of weeks. It taught me what delivering software with other people actually takes.
WHAT I LEARNED
At Wclub I moved from projects to a product. Code I wrote reached real people, so reliability, edge cases, and maintainability stopped being abstract ideas.
WHAT I LEARNED
At Wrtn, a generative AI platform, I worked on systems serving a large and growing user base. Scale changed how I think about design decisions, trade-offs, and performance.
WHAT I LEARNED
Today I focus on backend and software engineering: dependable systems with a human touch. The habits I picked up in that library classroom are still the ones I rely on most.
WHAT I’M FOCUSED ON
02EXPERIENCE
A place for roles, responsibilities, and meaningful engineering outcomes.
MOST RECENT FIRST
Crack image moderation pipeline & QA system
PROBLEM
Image review was handled manually. The workflow needed to automate repeatable decisions while preserving operator review and a traceable record of both AI and human verdicts.
APPROACH
Built a Python·FastAPI moderation API and asynchronous Inference Worker. AI models assessed NSFW, nudity, violence and related image risk, then Risk Scores and policy Thresholds routed each image.
OUTCOME
Human-in-the-loop QA compared AI verdicts with operators’ final decisions for False Positive·False Negative analysis and Threshold tuning.
PythonFastAPIPyTorchHugging Face TransformersDockerAWS S3SQSEC2
ViT/EVA 기반 NSFW ClassifierNudeNetViolence/Gore Classification Model
Internal document Q&A and faster onboarding for new hires
PROBLEM
The operations team depended on searching Notion manuals and asking senior colleagues, making it harder for new hires to find the source behind an answer.
APPROACH
Chunked and embedded internal documents into Qdrant. Semantic Search retrieved related documents and passed them to the LLM as context so staff could inspect supporting sources.
OUTCOME
30 days→< 7 days
Average ramp-up time for new operations staff was shortened, while reliance on senior colleagues also decreased.
PythonFastAPILangChainQdrant
OpenAI Embedding ModelsOpenAI API
Node.js 14→20 and core dependency migration
PROBLEM
The production stack was approaching EOL across Node.js 14, NestJS 7 and TypeORM 0.2, requiring a coordinated upgrade of core runtime and dependencies.
APPROACH
Solely led the migrations to Node.js 20, NestJS 10 and TypeORM 0.3. Audited about 80 repositories, then completed unit, payment E2E and staging verification before release.
OUTCOME
120ms→80ms
As a project outcome, key API SQL calls fell from 5 to 3 and p95 latency improved, as confirmed in ELK.
NestJSTypeORMDockerAWSELK
Automated cancellations & refunds, lower churn
PROBLEM
Subscription cancellations and refund requests were handled primarily by agents, creating repetitive phone work for the operations team.
APPROACH
Implemented customer Self-Service IVR scenarios for a one-month free extension or subscription cancellation, together with the related backend processing flow.
OUTCOME
18.5%→< 10%
The rollout reduced repetitive phone work and contributed to the drop in the cancellation rate after launch.
TypeScriptNestJSMySQLAWS ECS
ATALK BIZ IVR API
New PG integration and payment system improvements
PROBLEM
A recurring billing system handling about 100K payments a month needed a new PG integration, reliable event delivery and recovery from state mismatches. Daily cancellations were also handled manually.
APPROACH
Integrated Hecto PG, then applied Transactional Outbox and Reconciliation to payment processing. Connected the PG cancellation API to remove repetitive transaction lookups, cancellation requests and hand-offs.
OUTCOME
350–500 / day
Payment event loss could be prevented and PG–internal DB mismatches detected and recovered. Daily cancellation work was automated at the operating scale shown above.
TypeScriptNestJSMySQLAWS SQSECSDocker
Hecto Financial PG API
03RECOMMENDATIONS
Recommendations from teammates and colleagues who built software with Ryan M. Park.
SCROLL TO READ
I had the opportunity to work with Ryan in our Chingu Voyage. Ryan takes challenges proactively, and he is a very ambitious person eager to learn new technologies and improve his knowledge. While working with him, he constantly improved the code base and the project, having demonstrated technical skills such as responsive UX design and front-end development. Ryan knows how to make the best out of opportunities and the has the ability to motivate himself to deliver clean, efficient code based on specifications in a timely manner. Having a positive attitude and excellent communication skills, Ryan showed that he can be a valuable team player, always ready to help colleagues around him. I would gladly recommend Ryan for any software developer job as he would be an asset in any company.
