Junction 2023
Stainless Hill, a custom-GPT-based platform for predicting steel prices, news impact, and patent development.
This hackathon was my sixth experience in the Junction hackathon series, and my eighth hackathon overall. That year, my team and I decided to participate in the sustainability-related challenge proposed by Outokumpu.
The demo website is available here. Further details and materials can be found here.
Stainless Hill: Concept for Our Solution
We proposed the new “Stainless Hill” platform, which enables, scores, and validates reliable data resources for sustainability in the steel industry for Outokumpu.
The internal application presents users with a dashboard that provides data and forecasts in key areas: alloy price forecasts, news affecting markets and demand, and the latest emerging patents related to manufacturing and the steel industry more broadly.
Internal Engines of the Stainless Hill System
As data sources, we planned to use three different custom GPTs (introduced by OpenAI in November 2023). These would be built by subject matter experts and continuously updated. If these custom GPTs were humans, they would be the most knowledgeable people in their respective fields.
The data fetched by our system, along with the resources it uses, goes through a reliability check. Our reliability and relevance scoring system leverages AI-powered GPT models.
We call this agent GPT for Scoring (GSC). It provides the reliability score using advanced prompts that take several scores into account. These scores are provided by GPT for Selection (GSL). GSL selects how many recruited GPTs to use and takes user feedback into account, for example feedback provided in the OpenAI GPT Store. GSL is our custom GPT recruiter. It monitors the GPT Store, identifies candidates, and recruits the best custom GPTs in each domain of expertise based on predefined criteria, such as highest rated, most used, newest, or randomly selected within a field for testing purposes.
Resources that meet the minimum reliability score are delivered to application users. Although the AI-powered system works to provide the most reliable data possible, users can still review the reliability score before making any important decision. To support future improvements in the scoring strategy, GSC also includes a feedback system.
If you have any feedback or comments, please feel free to leave them in the section below. Thank you!