Kakao Reveals 27 Technical Sessions for Day 1 of if(kakao AI) 2024

Kakao has revealed the Day 1 schedule for the developer conference if(kakao AI) 2024. The event focuses on AI and data, consisting of 27 afternoon technical sessions that share various internal AI application cases and research achievements.

Key Tracks and Session Content

The sessions are divided into 7 tracks: AI Agent, AI Service, AI Platform, AI Native, Multimodal AI, AI Responsibility and Safety, and Data. (AI Service track added per original text)

• AI Agent: Covers Context-Aware AI Mate which understands user context, an internal RAG-based bot for Kakao's business efficiency, and application design techniques that minimize latency by analyzing LLM inference structures.
• AI Platform: Introduces the core functions (Store, Serving, Training, Agent) of the Kakao AI Platform (KAP) for enterprise-wide AI development and operations. It also shares the case of adopting Nvidia Triton for a message ad recommendation system serving over 500 models daily, and Kakao Pay's experience optimizing LLM serving frameworks.
• Multimodal AI: Reveals insights from Kakao's proprietary multimodal LLM development, selected as a highlight paper at CVPR 2024. It also includes personalized text-to-image (T2I) generation models and generative AI-based named entity recognition (NER) technology for medical data.
• AI Responsibility and Safety: Introduces Kakao Bank's proprietary LLM evaluation framework DUO, SLM fine-tuning for smishing detection, and graph-based malicious user group detection technology. It also covers efforts to ensure the safety and reliability of AI services.

Data and Infrastructure

In the Data track, strategies for real-time data processing and monitoring using Spark Streaming in AI-based ad recommendation pipelines will be shared. Use cases of MLOps on Kakao Cloud and video streaming with AMD accelerators will also be presented.