Daehong Communications, an advertising agency affiliated with Lotte Group, built D-AISY with Smart Developer so employees can use company-built AI services through their work accounts in one place. Core features include account and access management, a home screen, workflow and AI agent menus, and administrator analytics. Before the platform, collecting broadcast and OTT programming information took two days, followed by one to two hours of daily page checks to catch changes. Researching and organizing materials for advertiser meetings also took two days. Staff used external AI tools individually to create factbooks and social media copy, dispersing work context and materials outside the company.

D-AISY includes five task-specific AI solutions, including advertising report automation, program lineup organization, factbook support, and social media copy generation. The first release centers on a program lineup AI that checks for new programs and schedule changes on a set cycle, extracts information from schedules uploaded by users, and stores it by category. The platform also incorporates an in-development automation tool for daily YouTube reports. Workflow and agent menus let the company add new solutions, while an administrator dashboard centrally tracks service usage.

The platform records employee AI usage and statistics so the company can use actual usage data to decide on future features. Program schedule information is processed by AI and automatically checked before a staff member confirms it through a human-in-the-loop review. Routine research and schedule changes, however, are applied automatically once they pass validation criteria. Developers separated items requiring human review from those suitable for automation.

Because search and web crawling are central to the program lineup AI, development began with items requiring copyright and legal review. The initial model did not change its search terms and try again when its first search failed. Perplexity's tool-calling and reasoning performance fell short of expectations, so the team standardized on the GPT API. It addressed the lack of date filters by strengthening search conditions in the system prompt. Instead of retrieving pages directly from its server, the system gets URLs through official Naver and Daum APIs, then passes them to Gemini's URL Context feature to obtain page content. The team proposed data collection methods that the client's legal team could review and discussed them based on test results; this process took the most time.

The platform resolved cases where capitalization differences made the same service name appear to be separate platforms by standardizing major OTT names, including Netflix, Disney+, and Wavve, in Korean. It automatically checks the sources for programming and cast information and their consistency with existing data, while limiting searches to 20 major news outlet domains. Failed AI tasks can be retried up to three times. The system logs failures for individual items and saves schedules as a transaction, canceling the entire operation if an error occurs midway.

Daehong Communications is discussing a second phase of improvements. Proposals under consideration include sharing prompts and workflows, searching by department, job role, and tag, and letting administrators view participation by department. The company is also preparing AI tools for mockup image creation and competitive presentation support for the second half of the year. Developers said the hardest decisions involved the available data, collection methods that could pass legal review, and assigning staff to check erroneous results, rather than model or interface implementation.

Daehong Communications, an advertising agency affiliated with Lotte Group, built D-AISY with Smart Developer so employees can use company-built AI services through their work accounts in one place. Core features include account and access management, a home screen, workflow and AI agent menus, and administrator analytics. Before the platform, collecting broadcast and OTT programming information took two days, followed by one to two hours of daily page checks to catch changes. Researching and organizing materials for advertiser meetings also took two days. Staff used external AI tools individually to create factbooks and social media copy, dispersing work context and materials outside the company.

D-AISY includes five task-specific AI solutions, including advertising report automation, program lineup organization, factbook support, and social media copy generation. The first release centers on a program lineup AI that checks for new programs and schedule changes on a set cycle, extracts information from schedules uploaded by users, and stores it by category. The platform also incorporates an in-development automation tool for daily YouTube reports. Workflow and agent menus let the company add new solutions, while an administrator dashboard centrally tracks service usage.

The platform records employee AI usage and statistics so the company can use actual usage data to decide on future features. Program schedule information is processed by AI and automatically checked before a staff member confirms it through a human-in-the-loop review. Routine research and schedule changes, however, are applied automatically once they pass validation criteria. Developers separated items requiring human review from those suitable for automation.

Because search and web crawling are central to the program lineup AI, development began with items requiring copyright and legal review. The initial model did not change its search terms and try again when its first search failed. Perplexity's tool-calling and reasoning performance fell short of expectations, so the team standardized on the GPT API. It addressed the lack of date filters by strengthening search conditions in the system prompt. Instead of retrieving pages directly from its server, the system gets URLs through official Naver and Daum APIs, then passes them to Gemini's URL Context feature to obtain page content. The team proposed data collection methods that the client's legal team could review and discussed them based on test results; this process took the most time.

The platform resolved cases where capitalization differences made the same service name appear to be separate platforms by standardizing major OTT names, including Netflix, Disney+, and Wavve, in Korean. It automatically checks the sources for programming and cast information and their consistency with existing data, while limiting searches to 20 major news outlet domains. Failed AI tasks can be retried up to three times. The system logs failures for individual items and saves schedules as a transaction, canceling the entire operation if an error occurs midway.

Daehong Communications is discussing a second phase of improvements. Proposals under consideration include sharing prompts and workflows, searching by department, job role, and tag, and letting administrators view participation by department. The company is also preparing AI tools for mockup image creation and competitive presentation support for the second half of the year. Developers said the hardest decisions involved the available data, collection methods that could pass legal review, and assigning staff to check erroneous results, rather than model or interface implementation.