AI Briefing#kurly
The latest AI and developer news about #kurly, with the original source and a short summary.
The latest AI and developer news about #kurly, with the original source and a short summary.
The team completed work that would have taken 23 business days manually in 3.5 days, expanding the scope of regression testing.
Curry has built a daily automated deployment pipeline for recommendation models using Argo Workflows and SageMaker.
Shares the process of building an AI system that efficiently delivers domain knowledge by combining LLM Wiki and RAG.
Kurly has developed an On-device system that compares the similarity of delivery-completion photos to detect mis-deliveries in real time.
It uses the Outbox pattern and retries to safely synchronize external inbound-schedule information.
Kurly introduced a Collaborative Embedding model with Knowledge Distillation applied, achieving approximately 140% ARPU increase on sale pages.
Kurly reduced delivery-completion photo inspection costs by 93% using a YOLOv11-based object detection model.
Kurly developed an AccessBlock system that blocks service maintenance access at the application level without nginx permissions, and improved performance with Redis caching.
Kurly's OMS team secured stable order processing and deployment efficiency with a small workforce through a Shared Cache architecture and MSA separation.
Kurly integrated OMS and TAM to automate order distribution planning and improve operational efficiency.
Kurly's OMS team launched an internalization project for address refinement to cut up to 30 million KRW in monthly external costs and secure service stability.
Kurly internalized the address refinement logic for standalone buildings using its own building DB, reducing external vendor API calls by 60%.
Kurly internalized its multi-building address refinement logic, reducing external API calls and achieving 99.2% accuracy.
The Kurly team resolved an OOM issue caused by a connection leak that occurred after switching the MySQL connector, using Heap Dump analysis.
Kurly's Delivery Product Team organized indiscriminate error logs and set clear standards, reducing alarm fatigue and improving service stability.
Kellyz's Order Platform Development team combined Spring Kafka's ConsumerSeekAware with Redis Pub/Sub to shift offsets without stopping the consumer group.
Kelly completed a full switch from its SaaS-based delivery system to an in-house developed system, achieving zero failures during the transition.
Kelly used an AnyLogic-based digital twin to simulate the picking process and verify the effects of optimizing worker paths and inventory placement.
Kurly unveiled a method that uses LLM and Chain of Thought techniques to accurately and transparently curate unstructured review data without an existing ML model.
Kurly adopted Google Vertex AI Search to resolve typo and synonym issues, improving purchase conversion by over 300% in no-result (NR) search cases.
Kurly automated review summarization and analysis by calling Gemini Pro directly from BigQuery, without building a separate model.
Kurly acknowledged the limitations of CNN models and improved blurry image detection accuracy using a rule-based algorithm based on the Laplacian filter.
Kurly introduced motion to its quick menu icons and verified an increase in Kurly Members sign-up rate.
A Kurly developer pointed out the inefficiency of JPA saveAll and improved 1:N data processing performance using JDBC Bulk Insert leveraging Auto Increment PK inference.