professional-work

05

Foodservice e-commerce — maintenance and search evaluation

I maintained web and app interfaces for a food commerce service and connected customer search to a new API. I also built same-condition search comparisons, a test-set API and an admin for reviewing dictionaries and indexes.

preview
같은 테스트셋으로 두 검색 결과를 비교하는 설명용 도식이며 실제 평가 결과가 아닙니다 / Illustrative search ranking comparison, not measured results

Changing an existing service

In the foodservice e-commerce project, I contributed to maintenance of the live JSP/jQuery commerce site and its Android and iOS WebView apps. I changed promotion and order displays, aligned member lookup for identity verification, separated app environment settings and improved navigation to external links and app stores. The work started with understanding the existing flow and finding the parts that needed to change.

Search replacement was part of that setting. I implemented code connecting customer search to a new API and handled the search admin frontend, evaluation tools and test-set API. Other developers were responsible for the search engine and production search backend.

Deciding what to compare when search changes

Connecting an API does not establish that search has improved. We needed to compare the same queries under the same conditions and examine where expected products appeared in the rankings.

I built an evaluation screen that runs the old and changed search against the same test set. It shows individual cases alongside ranking metrics such as nDCG and MRR. I also used FastAPI and SQLite to store and edit test cases and judgments, with CSV import and export. This kept the conditions and judgments available for subsequent comparisons.

Connecting reviewed changes to operational settings

I built a separate operational admin. Saving a dictionary draft was distinguished from generating, comparing or applying an index. Same-condition simulations and ranking configuration APIs let operators review the effects of changes and inspect their application state.

I connected OpenAPI response validation and organized guides and integration test scenarios around the sequence of operational tasks. I used AI development tools and took responsibility for defining the problem, integrating the implementation, connecting APIs and establishing the verification flow.

My implementation produced customer search API integration code and evaluation and administration tools. The test-set API stored comparison conditions and judgments, while the production search backend was developed by others.

Search rules and criteria for useful results

In my personal Rhyme Generator assignment, searching by the last vowel exposed limitations in the definition and dictionary coverage. Reading it alongside this work now separates the rules that produce results from the criteria used to judge them. This is a question revisited in another setting, rather than a direct evolution from the assignment into the evaluation tools.