Jeonghyeon Lim

Projects

My main work starts with classroom needs: lesson planning, teacher review, and editable documents. In other projects, I explore how AI can help people review information and make decisions.

Main Projects

ZZAIM (짜임)

Lesson planningK–12 curriculumTeacher reviewEditable HWPX

An AI lesson-planning tool for Korean K–12 teachers. It connects a lesson idea to a draft, teacher review, and an editable HWPX document, a Korean word-processing format. I integrated HWPXKit’s generation engine so the download reflects the version the teacher approved.

From idea to classroom document

  1. Draft

    Describe the subject, grade, and topic to start a lesson plan.

  2. Review

    Review proposed additions or revisions and approve what to keep.

  3. Export

    Download the approved version as HWPX to edit in Hancom Office.

Open ZZAIMChatGPT Apps

HWPXKit

HWPXConversational editingDocument engine

A web tool for uploading an HWPX document, requesting edits in conversation, and downloading the revised file. It is also available through ChatGPT and Claude connectors. Its underlying engine supports reading, generation, and editing, and also powers HWPX generation in ZZAIM.

Editing a document on the web

  1. Upload

    Sign in and open an HWPX document in the browser.

  2. Request edits

    Describe what to change and check the changes for the current task.

  3. Download

    Download the revised file before closing the tab. Open it in Hancom Office.

Open HWPXKitChatGPTClaude

SemanticCompass (나침반) / ToolCompass

CurriculumFine-tuningTool selectionResearch

SemanticCompass explores whether a small model can supply useful domain context to another AI model. I released a Qwen3-4B LoRA adapter, a small set of fine-tuned weights, for mapping teachers’ questions to Korean curriculum codes and key concepts. The archived demo preserves examples; live inference is disabled. ToolCompass extends this inquiry to how language models select tools, with evaluation and reproducibility as ongoing research questions.

Archived demoHugging Face

Side Explorations

Winnow

AI-generated imagesReview orderHuman judgment

A tool that surfaces review signals in AI-generated images and orders a batch for review. It shows which signals affect the order while leaving the final selection to the user. The public demo presents recorded reviews of test images I created.

From an image batch to a review order

  1. Inspect

    Look for signals such as text issues, watermarks, and distorted anatomy.

  2. Order

    See which recorded signals move an image later in the review order.

  3. Decide

    Read the reasons and decide which images to use.

View review examples

Eodi

Knowledge baseRulesAI navigation

A personal project mapping airline and hotel loyalty programs into a structured, queryable knowledge base, built out of curiosity about how complex rule systems can be made navigable through AI.