LLM Structured Outputs: moving from Chat to the Useful & Hear about Trade Me's GenAI experiments
Topic 1: Models need interaction methods better than just text to help developers create more than the typical "Chat wrapper". OpenAIs 'Structured Outputs' leads the way with 100% accuracy. Google Gemini has code_execution that works well for arithmetical or logic problems. How can we use these APIs to create more substantial applications. Presented by Mick Clarke.
Topic 2: Developing an internal AI experimentation platform at Trade Me - Generative AI has the potential to create huge disruption, but also to be time wasting hype - and distinguishing between these is very challenging without experimentation. In order to identify the opportunities at Trade Me, we decided to develop an internal Generative AI platform that would allow us to experiment internally with AI such that we can raise our level of hands-on experience and identify viable opportunities at a relatively low cost. Presented by Calvin Giles.
I will demo the system I developed and the choices I made in creating this platform - based on FastAPI, Vertex Gemini, Streamlit and Google Cloud Run.
Kindly hosted by the good people at Trade Me.
Thanks Nicolo Musmeci.