Internal AI Desktop Tool

Problem

Internal teams needed a way to query and reason over private documents without sending data to external AI services, while still keeping the interface usable for non-technical users.

Constraints

• Data could not leave the local machine or approved internal environments
• UX had to be simple enough for non-technical stakeholders
• Unclear and evolving product requirements
• Limited time to validate usefulness before wider rollout

Approach

I designed and implemented a desktop UI that paired a local retrieval pipeline with a focused, minimal interface. I worked closely with backend and data science teammates to shape the UX around real usage patterns, iterating quickly based on feedback rather than over-engineering upfront.

Outcome

The prototype was successfully piloted internally and demonstrated that on-device retrieval could support real workflows. Stakeholders were able to test concrete use cases, which helped clarify both the value and the limitations of the approach.

What I’d improve

With more time, I would invest in better document ingestion tooling, clearer affordances around model limitations, and more robust evaluation metrics to guide future iterations.