Joonmo Jeong (i3months)

Developed STACK-UP - An AI Interview Platform for IT Roles

• GitHub Repository
• Served as development lead and AI server developer - built AI modules (RAG, LangChain, VectorDB, Embedding) and designed the software architecture.
• Boosted development productivity by actively leveraging LLMs such as Claude Code.
• Applied state-of-the-art prompt engineering techniques to the evaluator personas and scoring criteria to secure output consistency and evaluation reliability.
• Designed multi-interviewer panel feedback (multi-agent)
• Calls a panel of evaluators in parallel via prompt multi-calls on the same LLM differing only in persona and evaluation axis - technical (technical_accuracy), logic (logic_score), and communication (using voice metrics, communication_score) evaluators (swapped for personality/collaboration evaluators in PERSONALITY mode).
• Computes the overall score via code-based weighted averaging (0.5·technical + 0.25·logic + 0.25·delivery) and generates narrative strengths/weaknesses and a study_plan with a single synthesis LLM call.
• Adaptive question generation that follows up topic by topic - questions and follow-ups use a single LLM, while multi-agent is applied only at the feedback stage.
• To mitigate scoring noise, designed the flow as per-answer evaluation (specificity, logic, structure, correctness) → deterministic aggregation → LLM final judgment, attaching rationale that pinpoints where points were deducted.
• Built a RAG pipeline that chunks and embeds job/technical documents into a VectorDB and verifies the factual accuracy of candidate answers using retrieved context chunks.
• Report deliverables - provides scores, narrative evaluation, rationale, and user history management, along with a shareable structured report and PDF export.

Autonomous Maze Navigation Robot Based on Ultrasonic & Infrared Sensors - Embedded Challenge

• GitHub Repository
• Built a maze-navigating autonomous robot on STM32F429 (Cortex-M4) and FreeRTOS, fusing three ultrasonic and two infrared sensors to converge on the north exit without any precomputed global path.
• Rejected a full-stack Occupancy Grid + A* path planner due to hardware constraints (left/right motor output asymmetry, absence of an IMU), reducing the design to a robustness-first approach that hardcodes the map (revealed on the day) and probabilistically models only the dynamic-obstacle segment.
• Greedy Cardinal FSM northward navigation - discretely detects corners and intersections via lateral-wall distance change (Δd) and branches on cardinal heading (relative direction), making it robust to accumulated position drift.
• EKF sensor fusion design
• Accumulates wheel odometry with RK2 (midpoint) integration, and computes the measurement Jacobian in closed form via analytical ray-segment intersection with the hardcoded wall segments for the ultrasonic sensors.
• Per-sensor sequential 1D Kalman updates with innovation gating (rejecting |y| > 3√S) and Joseph form covariance updates to prevent divergence and preserve symmetric positive-definiteness.
• Anchors heading drift caused by the missing IMU to an absolute reference via an arctan heading observation from lateral-wall geometry (θ = θ_wall + arctan(Δd/L)).
• Dynamic-obstacle probabilistic modeling - time-weighted accumulation and decay of occupancy probability at sensor beam endpoints separates static walls from moving objects, and temporarily boosts cruise speed (V_cruise → V_dash) to pass quickly once a passage window opens.
• Extensive calibration - compensated battery dependency and left/right motor asymmetry with separate left/right substeps and closed-loop pivot, and implemented LED priority-encoded debugging to read runtime state without a serial connection.
• Separated the sensing and control layers with a four-task FreeRTOS design (Sensor, IR, Control, Debug) under priority preemptive scheduling and lock-free single-producer/single-consumer shared state.
• Practice run (full completion) video

PACK-UP - Developed and Operated a Travel Booking Platform

• packup.asia
• Personally drove the entire process from planning to development, deployment, and operation - a production-ready service with both a user app and an operations admin, in service after launching on Google Play and the iOS App Store.
• Studied the official Claude Code documentation and built and used custom skills tailored to the project, maximizing development productivity in an AI-friendly way.
• System Composition
• Mobile App: React Native + Spring Boot
• Admin: Spring Boot + React
• Separated into multiple servers - core, payment, chat, batch, and admin.
• Deployed on GCP Cloud Run.
• Architecture Design & Considerations
• Adopted Feature-Sliced Design (FSD) architecture in the React Native app for a feature-based modular structure, ensuring maintainability and scalability.
• Placed Cloud Armor (GCP WAF) between the load balancer and application servers to block known attack patterns such as SQLi and XSS, and applied per-IP Rate Limiting (429 on exceeding 300 req/min) and ML-based DDoS detection via Adaptive Protection.
• Reviewed a CQRS design separating write (Command) and read (Query) logic with JPA + MyBatis, considering store separation across Elasticsearch, Redis, and PostgreSQL.
• Examined MSA from an AI-friendly architecture perspective
• A large monolith is hard for AI to grasp in full context, whereas MSA lets AI agents take charge of a domain on a per-service basis.
• As model context windows grow, a monolith may become understandable without splitting it - so the criteria for service separation warrant continued consideration.

Developed UniScope - A Platform for Nationwide University Information

• GitHub Repository
• Kakao Tech Campus Final Presentation - UniScope Platform
• Documented UI components across the project using Storybook to enhance reusability and improve development speed and design consistency.
• Built an E2E testing environment using Playwright and automated actual browser-based UI flows and key business logic.
• Improved page structure and performance to achieve high scores across 4 Lighthouse metrics, and enhanced initial accessibility through meta tag-based SEO optimization.
• Applied Husky at the pre-commit stage to enforce linting rules and built a CI/CD pipeline based on GitHub Actions.
• Introduced Feature Sliced Design (FSD) architecture and established related team guidelines to ensure maintainability and scalability
• Designed a deployment architecture integrating AWS and Vercel, and built a Blue/Green zero-downtime deployment environment using Nginx reverse proxy.

Developed features for the Chemical Information Platform Service

• Migrated data from TIBERO to MariaDB.
• Indexing migrated data and implemented partitioning for statistical data.
• Optimized queries from TIBERO to MariaDB.
• Implemented hierarchical queries and sorting using WITH RECURSIVE and SORT_PATH.
• Refactored high-cost statistical queries, improving performance (873.231ms → 214.245ms).
• Visualized statistical data using ApexChart and implemented favorites and table search features.

Enhanced functionalities of the DBpia service

• Integrated Redis for floating banners and developed APIs to connect the back office with the DBpia service.
• Modified Elasticsearch queries and existing mappings for the unified search feature, restructured the research paper index.
• Analyzed execution plans, reindexed data, and optimized high-cost queries (4481.926ms → 1111.281ms).
• Managed encrypted view schemas and handled encryption/decryption of sensitive information.
• Maintained DBpia service operations – Fixed Excel download bugs, managed the PDF download server, and handled long-term inactive users.

Developed DBpia Content Maker (DCM)

• Analyzed legacy source code written in C# and C++ and designed a new web-based system.
• Utilized Git Flow strategy for seamless collaboration, analyzed user requirements and system processes.
• Set up and managed project configurations and operations.
• Configured Spring Security with custom filters and authenticators, implemented LDAP-based login.
• Customized Spring Security authentication exception handling - Implemented account lockout policy.
• Resolved DataSource TLS compatibility issue with SQL Server 2005.
• Fixed incompatibility between the latest Spring Batch version and SQL Server 2005 sequences.
• Established certificates and domains, NGINX-based reverse proxy routing for development and production servers
• Deployed and managed Tomcat on Windows Server, resolved symbolic link-related network drive permission issues
• Configured Tomcat JVM memory pool and implemented redundancy for production and development servers.
• Handled penetration testing issues, including XSS, CSRF, and sensitive data masking.
• Analyzed and optimized SQL Server Linked Server RPC overhead using OPENQUERY.
• Research Paper Metadata Production System Development
• Developed PDF splitting system (PDFBox, PDF.js) - Handled logical/physical page differentiation
• Refactored a legacy C++ batch system into Spring Batch, integrating ERP and DCM services to enhance stability and processing speed.
• Developed a temporary storage system for research papers to periodically save input data.
• Implemented an email-based research paper author search feature using Elasticsearch.
• Efficiency Improvements & Service Optimization
• Revamped the research paper production system, reducing metadata processing time (30 min → 15 min for 50 papers).
• Optimized batch processing, cutting publication time (3 days → 1 day).

Developed the ASD Screening AI Integrated Platform System

• Implemented conditional branching surveys using SurveyJS.
• Transmitted video and audio data via WebSocket for real-time communication with the AI inference server.
• Visualized inference results in real-time using ApexCharts and D3.js.
• Set up system execution environment and deployed images using Docker.
• Developed JWT authentication and CRUD functionality for participants and test results using FastAPI.
• Built and developed the frontend using Vue.

Enhanced functionalities of the ETRI School website

• Unified non-registered student and faculty data into a single table.
• Implemented bulletin board pagination and step-by-step registration using Vue Quasar.
• Developed a Spring Boot and JPA-based back-office system and integrated with the ETRI School service.
• Deployed and managed the system on ETRI's internal servers using Docker.

Developed the Drone Mission Data Management System

• Developed batch business logic for drone data storage.
• Traversed directories using DFS and stored image metadata.
• Detected changes at the project level and updated only modified data.
• Used AtomicBoolean api to ensure only one scheduler runs at a time.
• Handled scheduler tasks asynchronously with CompletableFuture api.
• Conducted performance benchmarking tests considering JVM warm-up.
• Visualized image data on a map using React.
• Deployed Frontend, Backend, and Database as a unified system using docker-compose.

Enhanced functionalities of the National Supercomputing Center website

• Redesigned the main page and developed the back-office system.
• Added an email verification step for user registration, including security measures and expiration settings for verification codes.
• Visualized supercomputer usage statistics using D3.js and optimized queries (6376.21ms → 430.71ms).
• Implemented a WebSocket-based broadcast feature for connected users.

• Added eziwiki to the Awesome Static Generators list
• Contributor
• https://github.com/myles/awesome-static-generators/pull/221
• Github Stars:

• A modern, lightweight wiki and documentation site generator built with Next.js
• Markdown-based content creation with customizable navigation
• Hash-based URLs for privacy protection and static site deployment support
• Owner & Maintainer
• Next.js with TypeScript
• https://github.com/i3months/eziwiki
• https://eziwiki.vercel.app/
• Github Stars:
• Codacy :

ABC Bootcamp: Data-Driven Course

• Developed "D-Village" a survival-based community building platform for small groups using React and FastAPI.

Chungnam National University Algorithm Competition – DevDay

• University-wide algorithm competition held every spring and fall.
• Bronze Prize in November 2025, June 2025, and June 2022 / Encouragement Prize in November 2021.

Chungnam National University Generative AI Challenge

• Developed "PPTree" an automated lecture material generation tool using LangChain.

Kakao Tech Campus 2nd Stage Outstanding Cookies Selection

• Completed the 2nd stage with the highest score in the Frontend track.

11th NTIS Information Utilization Competition

• Developed "NTIS STUDIO" a scientific document editor using Vue and FastAPI.

K-Digital Challenge: 2023 MyData Idea Competition

• Developed "Trippy" a personalized travel app using Flutter.

2023 Environmental Data Utilization and Analysis Competition

• Planned "CEAP" a platform for analyzing company-specific carbon emissions using greenhouse gas data.

Chungnam National University Junior Creative Works Competition

• Developed a refrigerator management service using deep learning-based image recognition.

Chungnam National University

Chungnam National University Practical Coding Course Tutor

University of Nevada, Las Vegas (UNLV) - Visiting Scholar

Seoul National University Next-Generation Semiconductor Innovation Convergence College - Undergraduate Research Intern

KakaoTechCampus 3rd Cohort - Frontend

AI Career School Startup Hackathon L:AUNCH Completion

Chungnam National University 2025 SW-IT Contest Organizer

• Baekjoon Online Judge (BOJ)
• solved.ac
• Codeforces
• i3months-web-shell

TOPCIT Regular Assessment

TOEIC Speaking

43rd Advanced Data Analytics Semi-Professional (ADsP)

54th Structured Query Language Developer (SQLD)