KR

AI-Driven Problem Solving Portfolio

Five product cases showing how I automate repetitive work, structure unstructured data, and turn outputs into actionable insights.

Shared Problem, Shared Approach

Problem Pattern

Across all five projects, information is unstructured, scattered, and hard to reuse.

Solution Pattern

I normalize user input and external data, then connect AI outputs to concrete actions.

Impact Pattern

This reduces repetitive work, improves decision speed, and compounds personal knowledge assets.

AI Usage Principles (Summary)

Collection → Normalize → Infer → Serve → Improve
  • Structure inputs first, then apply AI to analysis and inference.
  • Connect model outputs to UI, storage, and feedback loops.
  • Measure impact by how much repetitive user work is removed.

AI Problem-Solving Flow (Across 5 Cases)

1. Problem FramingDefine repetitive, missing, and inefficient tasks
→
2. Data CollectionCollect text, records, and external inputs
→
3. NormalizationSchema mapping, field extraction, structuring
→
4. AI InferenceSummarization, extraction, analysis, evaluation
→
5. Serving & FeedbackUI integration, storage, and re-analysis loop

Case Studies

Case 01. Blog Knowledge System

data-droid.github.io
Solution

A Jekyll-based technical blog focused on data engineering topics.

Problem

Learning and work notes were scattered and difficult to reuse.

AI Application

Used AI for draft generation, structuring, and style refinement.

Model
LLM | Draft Generation LLM | Summarization LLM | Style Refinement
Result

Publishable drafts and structured technical articles.

Key Contributions
  • Built an archival writing structure to compound reusable knowledge.
  • Standardized article flow to reduce writing and publishing lead time.

Case 02. CareerWeb Automation

CareerWeb CareerWeb-backend
Solution

A workflow web app for job-post collection, Notion sync, and ATS analysis (React + FastAPI).

Problem

Manual job-post review and comparison was repetitive and time-consuming.

AI Application

Used AI for job parsing, ATS fit analysis, and structured report generation.

Model
Gemini | Job Posting Parsing Gemini | ATS Fit Analysis Gemini | Report Generation
Result

Structured job data and ATS feedback reports.

Key Contributions
  • Converted manual collection-to-analysis steps into an automation pipeline.
  • Improved reusability through persistent Notion-based analysis records.

Case 03. sceneNote Learning Intelligence

sceneNote
Solution

A SwiftUI subtitle-learning core (`SubtitleCore`) with continuous study/review flow.

Problem

Expression collection and review loops often broke during subtitle-based learning.

AI Application

Used AI for subtitle chunk analysis, expression extraction, and learning-ready processing.

Model
Gemini | Subtitle Chunk Analysis Gemini | Expression Extraction TTS Model | Pronunciation Playback
Result

A reusable expression list and pronunciation-based review workflow.

Key Contributions
  • Modularized core learning logic as SPM for maintainability and reuse.
  • Connected extraction to review UX for better learning continuity.

Case 04. coffeeJournal Data Capture

coffeeJournal
Solution

An iOS app for beans, recipes, cups, and brew-history tracking.

Problem

Missing brew records made taste reproducibility difficult.

AI Application

Used AI-assisted extraction and mapping to reduce manual record entry.

Model
Gemini | Text Extraction Gemini | Field Mapping Gemini | Entry Assistance
Result

Auto-assisted bean and brew record data.

Key Contributions
  • Integrated bean/recipe/brew logs into a single reproducibility workflow.
  • Reduced input friction with AI-assisted entry automation.

Case 05. RecipeBook Structuring Engine

recipeBook
Solution

An iOS recipe app supporting save/search and cooking-mode execution.

Problem

Unstructured online recipes were hard to convert into reusable personal assets.

AI Application

Used AI parsing to normalize recipe text and automate content extraction.

Model
Gemini | Recipe Parsing Gemini | Data Structuring Gemini | Content Extraction
Result

Reusable, structured recipe data for downstream use.

Key Contributions
  • Improved search and reuse by converting unstructured recipes to structured data.
  • Increased usability by separating cooking-mode and detail-view flows.