developer@lunark

We’re building KAIdol, an emotionally immersive character chatbot platform designed to simulate deep human-like relationships through K-POP inspired AI idols. Our interests span across: Multilingual Emotion Recognition (12+ fine-grained emotions, real-time inference) Intent Detection with lightweight BGE-M3 & Gemini Classifiers Hybrid Memory Systems combining Redis, Firestore, Qdrant, and Vertex AI Memory Bank Agent-Oriented LLM Orchestration using Vertex AI ADK + LangChain for memory, emotion, and relationship-aware conversations Closed LLM Tuning Strategies (Gemini, Claude, OpenAI) through context engineering, RAG, tool calling, and system prompts Character Persona Management & Prompt A/B Testing via JSON & Opik On-device LLM Experiments (Phi-2, Gemma) for local emotion/intent prediction Gamified Business Models integrated with in-chat gifts, point systems, and payback mechanics Our stack focuses on cost-efficient LLM routing, real-time affective response, and memory-driven persona consistency across multilingual audiences (KR/EN/ES). We're scaling toward 1.6M MAU with global expansion into North/South America & Korea. Let’s connect if you’re interested in emotion-aware agents, LLM orchestration, or generative fandom economies!