📋 Project Overview
Developed during internship at CJ AI Center, this project creates a sophisticated Multi-Agentic pipeline for analyzing Korean drama scripts using LangChain.
🎯 Problem Definition & Goals
- Problem: Analyzing drama scripts manually is time-consuming and subjective.
- Goal 1: Build an automated pipeline for extracting structured information from scripts.
- Goal 2: Implement multi-agent architecture for specialized analysis.
- Goal 3: Create a scalable system for batch script processing.
⚙️ Key Features & Contributions
- Multi-Agent Architecture: Designed specialized agents for character, plot, and emotion analysis.
- LangChain Integration: Built robust agent orchestration with custom tools.
- Korean NLP Processing: Implemented Korean-specific text processing.
- Structured Output: Generated JSON-formatted analysis results.
- Batch Processing: Designed pipeline for processing multiple scripts.
🔧 Technical Challenges & Solutions
- Script Format Variability: Developed robust parsers for various formats.
- Agent Coordination: Implemented shared memory and message passing.
- Context Length Limits: Created sliding window approach with intelligent chunking.
- Consistency in Analysis: Implemented entity resolution and character tracking.
📈 Results & Learnings
- Automation Achievement: Reduced script analysis time from hours to minutes.
- Analysis Quality: Produced valuable structured insights.
- Internship Contribution: Delivered working prototype demonstrating LLM applications.
- Key Learning: Gained experience in production agentic AI systems.