Jongjin Park

Backend Engineer · AI Applications

About Me

I have worked on Spring Boot and PostgreSQL backends and Python-based AI integrations. In the AI Trading Coach capstone, I was PM and worked on the order-decision backend, connecting user rules, model signals and mock orders. Model signals can exclude candidates, while rule-based code determines order size and final approval.

I maintain a fork of comic-translate, where I worked on local OCR and translation runtimes, the OCR request path and caching for repeated work. In a stock trading simulation, I built Spring Boot flows for orders, wallets and holdings.

I self-host Kavita, Jellyfin and personal projects on a NAS and OCI. I check changes with tests and run records, and record the input and comparison conditions alongside performance results.

Skills

AI / LLM

  • RAG: PostgreSQL and pgvector storage, access-aware retrieval, citation and number checks
  • Local OCR and LLM: connected PaddleOCR-VL, HunyuanOCR and Gemma runtimes through Docker and llama.cpp, with stage caching
  • Team AI server: wrote Docker, GPU setup and switching scripts for vLLM, Ollama and GPT-SoVITS

Back-end

  • Primary: Java, Kotlin, Spring Boot, Python, FastAPI, PostgreSQL
  • REST APIs, SQL, transactions, locking, database constraints, error and state handling
  • Also used: MySQL, Redis, WebSocket/STOMP

Infrastructure

  • Docker, Linux, GitHub Actions, Nginx, personal NAS and OCI services
  • Windows setup and run automation, pytest and JUnit, reproducible benchmarks

Others

  • C++ (including C++20): used mainly in university coursework and coding test practice
  • Kotlin Android, Jetpack Compose and Room project experience
  • Git, GitHub pull requests, technical documentation and cross-project API contracts

Experience

document-translate-myfix

Personal open-source fork · local OCR and translation runtimes

Apr 2026 – Present

GitHub
  • Implemented Docker start, stop and resource-release paths so OCR and translation models could use the GPU in turn.

Screening Humanity

Team prototype · FastAPI back-end / local AI server

Jan 2026

AI Server GitHub
  • Changed chat and speech APIs and the vLLM, Ollama and GPT-SoVITS setup and switching scripts.
  • GPU memory limits required choosing either vLLM or Ollama at a time. The final demo used the Ollama path.

Personal NAS and OCI Self-Hosting

Personal infrastructure operations

Winter 2022 – Present

  • Self-hosted a Kavita library, Jellyfin media service and personal projects with Docker on a NAS and OCI Linux VM.
  • Page turns in remote Kavita reading were delayed, so I set up image and PDF caching and next-page preloading. When preloading failed due to a certificate-name mismatch, I aligned TLS SNI and HTTP Host and checked response, cache and queue records.
  • Tested XPEnology on Proxmox, then removed the Proxmox layer. XPEnology is now the main NAS, with services running in Docker.

Projects

AI Trading Coach with User-Defined Risk Rules

Three-person capstone · PM / Decision Platform

May 2026 – Oct 2026

GitHub
  • Requiring both the rules and model to say BUY left over half of offline sessions without a candidate. I changed the flow to rule-proposed candidates with a model SELL veto.
  • Compared six combinations on the same 31 stocks and 5,362 sessions: no-candidate rate fell from 51.8% to 1.4%. I implemented the rule and added regression tests.
  • Mock-order tests covered submission, fills, balance updates and duplicate prevention after user and safety checks.
  • Separated financial sources, revisions, chunks and embeddings in PostgreSQL, with public/private retrieval and citation checks.
  • Teammates owned LSTM training, the Return Engine and dashboard UX.

Document and Image Translation Pipeline

Personal open-source maintenance fork

Apr 2026 – Present

GitHub
  • Added local HunyuanOCR, PaddleOCR-VL and Gemma paths to comic-translate.
  • Direct model requests cut OCR wall time from 60.2 to 8.3 seconds on the same six Japanese pages and 73 text regions; the final normalized OCR results matched.
  • Stage caching reuses matching input, model and settings. A two-page first run plus identical rerun took 279.9 seconds without caching and 156.0 seconds with it.

U.S. Stock Trading Simulation

Two-person team · Back-end / deployment

Jan 2026

Backend GitHub
  • Built Spring Boot flows for orders, wallets, holdings and trade history. Price requests run outside DB transactions; balance and holding changes use locking and transactions.
  • Cached prices with Redis and delivered them through WebSocket. I integrated a FastAPI AI server and deployment setup; a teammate owned the interface and design.

KB AnsimVoice and Money Rules

Three-person contest prototype · Agent / TTS

Jul 2026 – Aug 2026

Public GitHub
  • Rechecked the amount, recipient and account in an AI-drafted transfer against the user’s utterance and server data. Mismatches are not passed to the transfer flow.
  • Rule-based code handles transfer decisions and one-time approval. I worked on the constrained agent, input checks and TTS, WebSocket and Docker integration.

Net Zero Smart SCM

Individual database course project

Dec 2025

GitHub
  • Built a Java, Spring Boot and PostgreSQL course project for supplier, order, delivery, inventory and carbon data.
  • Kept order, delivery and inventory changes in one transaction, with database constraints and error handling to prevent inconsistent updates.

OliveME Personal Color Android App

Two-person team · Android architecture / data / integration

May 2026 – Jun 2026

Team GitHub
  • Built Android screens and storage flows with Kotlin, Jetpack Compose and Room, and connected AI and map APIs.
  • Kept saved diagnoses and local guidance available when an external API or backend fails. A teammate shared the screen design work.

College Student Flea Market

Nov 2025 – Dec 2025

GitHub
  • Built session authentication, product and comment flows, search and account pages with Express, EJS and MySQL in a two-person course project.

FPGA Morse Code Transceiver

Nov 2025 – Dec 2025

GitHub
  • Connected Verilog state machines, clock division and transmit/receive modules to buttons, LEDs, seven-segment displays and a buzzer on an FPGA board.

Busan Tourist Traffic Accident Analysis

Jun 2025

GitHub
  • Combined visitor, transport facility and accident data around 50 Busan attractions, then published statistical analysis, maps, a notebook and a report.

External Activities

KEEPER Security Club

Member

Mar 2025 – Aug 2026

  • Joined the web vulnerability study from Mar to Jul 2025.

APPTIVE

Member

Mar 2026 – Sep 2026

Kakao Tech Campus

Agentic AI · In progress

May – Nov 20, 2026 (expected)

  • Completion and project links will be added after verification.

Education

Pusan National University

Computer Science and Engineering

Expected graduation: Feb 2027