United StatesEducation2024 – ongoingUniversity of California, San Diego
TritonGPT administrative assistant and job-description writer
UC San Diego tested TritonGPT with approximately 500 early users before a planned expansion
Problem
Administrative staff needed help finding university-specific information and drafting documents. The university wanted an assistant hosted on its own infrastructure.
Approach
UC San Diego built TritonGPT using Meta's Llama 2 model and indexed university websites. Answers included supporting university sources. A job-description function drafted text from hiring managers' requirements for subsequent editing.
Results
Takeaway
Local hosting and cited university sources supported an institution-specific assistant. Employees remained responsible for checking generated answers and documents.
Categories
- Size
- Large enterprise
- Industry
- Education
- Business function
- Internal work and knowledge managementHuman resources
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
UC San Diego described TritonGPT's locally hosted architecture and university-source references. Approximately 500 early users tested the assistant, including drafting and summarization tasks, before a second pilot phase.
Full details are in the original sources.