metadata
pretty_name: Dongwon Agent-based Persona Simulator
tags:
- marketing
- simulation
- persona
- agent-based
- demand-forecasting
- time-series
- competition
language:
- ko
- en
license: cc-by-nc-nd-4.0
task_categories:
- tabular-classification
- time-series-forecasting
MiriMarket — LLM-based Consumer Persona Simulation
Dongwon X KAIST AI Competition (Award-winning Project)
🔍 Overview
MiriMarket는 새로운 제품을 출시하기 전에
실존 소비자들이 어떤 반응을 보일지 LLM 기반 가상 소비자(Persona) 시뮬레이션을 통해
수요를 예측하는 시스템입니다.
Dongwon X KAIST AI Competition에서 개발되었으며,
제품 기획 / 마케팅 / 타겟 세그먼트 전략 수립을 위해 설계되었습니다.
✨ Key Features
- LLM 기반 가상 소비자 생성 (Adult / Health-conscious / Convenience / B2B 등)
- Persona-specific 시장 반응(선호/구매의향/Flavor 선택) 시뮬레이션
- Monte Carlo 기반 다회 샘플링
- Bass Diffusion 기반 조기 수요 예측
- Streamlit 대시보드 가능 (선택)
🧠 System Architecture
Product Idea
│
├── PromptManager
│ ├─ segment-specific persona templates
│ └─ product briefing → persona reaction
│
├── Persona Generator (LLM)
│ └─ synthetic consumers (100~200 samples)
│
├── Persona Reaction Simulator
│ ├─ flavor preference
│ ├─ purchase intention
│ └─ NPS/emotional response
│
└── Market Forecast Engine
├─ Monte Carlo sampling
└─ Bass diffusion (optional)
📊 Example Output
- Persona distribution
- Flavor preference prediction
- 시나리오별 수요 변화 그래프(옵션)
📁 Repository Structure
/src
/persona_generator
/prompt_manager
/simulator
/notebooks
- analysis.ipynb
/data
- sample_personas.json (경량 샘플)
/README.md
📌 License
CC-BY-NC-4.0 (비영리 사용만 허용됩니다)
🏆 Awards
Dongwon X KAIST AI Competition — 장려상 수상
팀장 역할 + 전체 파이프라인 제작