Superb AI automates LH defect classification, analyzes 4 million images a year

[Financial News] Superb AI is using artificial intelligence (AI) to improve the efficiency of handling defect complaints in multi-unit housing and maintaining facilities. Automatically classifying defect images will reduce the workload for customer service staff, while accumulated data will be used to plan facility repairs and replacements. The move is seen as an effort to expand the use of AI-based data into construction and housing management.
According to industry sources on the 2nd, Superb AI has completed the first phase of a project to build an automated defect-classification system for LH based on a vision-language model (VLM), and is currently carrying out a second-phase upgrade.
LH has faced limitations with its existing classification method, which relies mainly on text and phone consultations, as it receives around 300,000 defect complaints each month. The classification criteria span five dimensions—space, component and material, defect type, construction work type, and project type—creating more than 18,000 possible combinations. In addition, judgments can vary depending on a counselor’s experience, making it difficult to maintain consistent classification quality.
In the first phase, Superb AI consolidated the defect-classification process, previously divided into multiple stages, into a single VLM model. The system analyzes each image and classifies all five categories at once, providing the processing result and the basis for its decision in natural-language sentences in real time, 24 hours a day, immediately after a complaint is submitted. One of its key advantages is that there is no need to operate separate models to locate defects and classify their types.
The system classifies major defect types, including cracks, leaks, damage, mold, and contamination. It also makes inferences based on learned patterns when the cause is difficult to identify from a photo alone. The system was designed with a target detection accuracy of at least 90% and false-positive and false-negative rates of no more than 10%.
As a result, customer service staff are expected to be able to focus on handling complaints that require judgment rather than simply receiving and classifying them. Structured defect data is also expected to help analyze patterns by housing complex, material, and years since construction. It could also be used to develop long-term maintenance plans, including to inspect locations where defects recur and determine when aging facilities should be replaced or repaired.
Superb AI CEO Kim Hyun-soo said, “This first-phase project is an example of resolving two challenges—complex classification criteria and large volumes of unstructured data—with a single VLM model.” He added, “We will carry out the ongoing second-phase upgrade without a hitch, while expanding the application of vision AI beyond defect classification into areas such as facility inspections and complaint image analysis.”
[email protected] Jang Min-kwon Reporter