Seity
A PoC of an enterprise AI chat where users check and edit sensitive information before it is sent to the AI. The PoC was completed, and the team signed an MOU with the security company Securelink.
Role
As team lead and frontend developer, coordinated the development schedule, mentoring and meetings across the BE, FE and AI parts
Problem
When a company uses generative AI, sensitive information such as names and addresses risks being sent to an external LLM as it is
Solution
- Built an interface that maps the positions returned by a sensitive information detection model onto the input box to highlight and mask them, and lets the user check and edit before sending
Result
Completed the PoC, and the team signed an MOU with the security company Securelink
Overview
Seity is a PoC built so that, when a company uses generative AI, the sensitive information contained in the input can be checked and corrected. In the SW Maestro 14th cohort project, as team lead and frontend developer, coordinated the development schedule, mentoring and meetings across the BE, FE and AI parts, and built the input review interface.
Key features
- Input review: Highlights and masks, in the input box, the positions of names or addresses found by the model
- User confirmation: The user looks over the flagged sensitive information, edits it directly, and then confirms what is sent to the AI
Implementation and improvements
-
Interaction that connects model results to the input box
The sensitive-information detection model, which analyzes the input sentence, returns the positions where sensitive information such as names or addresses appears. On the frontend, those positions are emphasized in the input box and the content is masked. The flow does not treat the automatic detection result as final. The user reviews and edits the flagged content and then sends it to the AI.
-
Coordinating product direction, development scope and schedule
As team lead, organized the product direction and the PoC scope, and coordinated the development schedule, mentoring and meetings across the BE, FE and AI parts. Matched the implementation scope of each part to connect the input review flow into the PoC.
-
Analyzing corporate privacy requirements
Signed an MOU with Securelink and analyzed the privacy requirements that companies have when using generative AI. This gave a more concrete view of the input review problem the PoC would address and of what a corporate environment requires.
Tech stack
Next.js · TypeScript · Tailwind CSS · React Query · Zustand
FastAPI
Results and current status
Completed a PoC for reviewing and editing sensitive information before it is entered into generative AI, and the team signed an MOU with the security company Securelink. My contributions are: coordinating the product direction and the schedule, mentoring and meetings of the BE, FE and AI parts; developing the input review interface; and analyzing corporate privacy requirements through the MOU with Securelink.