.svg)
.svg)
.png)
Pump helps their customers save 60% on AWS costs via group buying discounts & AI. In order to help their customers easily understand their AWS cost saving options, they developed an embedded co-pilot chatbot called PumpGPT. This allows their customers to ask AWS specific questions and discuss cost saving options with a chatbot directly on their platform.
When developing PumpGPT, it became clear that GPT-4 often hallucinated and generated inaccurate information because it lacked a specific real-world understanding of AWS cloud services, their cost structures, and the cloud ecosystem as a whole. While GPT-4 and other modern LLMs are powerful models, they often lack domain-specificity and depth leading them to hallucinate facts and information. Additionally, the original model would go stale as it struggled to keep pace with the ever-evolving landscape of AWS.
To help GPT-4 give more accurate and reliable answers, Pump needed to incorporate important data and context to help the model understand its complex domain-specific ecosystem. This means determining the essential data across both internal and external sources that were necessary for PumpGPT to offer accurate AWS information and relevant cost-saving recommendations – discovering and procuring this data, keeping it up-to-date, integrating it into the model and validating that it performs reliably for their product.
Using Arcus’ Prompt Enrichment product, Pump was able to discover and integrate internal data they already had that was valuable to PumpGPT, as well as, discover relevant external data and context that was necessary for their application. Pump used Arcus to seamlessly integrate this additional context into the prompts and augment the model used for PumpGPT, so that it could provide accurate information to their customers and stay up-to-date on the latest AWS information. Through this, the domain-specific data is continuously and directly integrated into the model to be served. By continuously augmenting the model’s data with the right datasets related to Pump’s product, AWS usage patterns, pricing, optimization strategies and much more, PumpGPT gained valuable insights, enabling it to create more accurate, personalized and actionable recommendations as well as prevent hallucinations and stale responses.
.png)
Accurate and Personalized Cost-Saving Recommendations:
With Arcus's Prompt Enrichment offering, PumpGPT showcased a significant improvement in its ability to give accurate and personalized cost saving strategies. By discovering and integrating the right domain specific data, it prevented their model from hallucinating and instead enabled it to provide factual and relevant responses, enabling customers to identify tailored cost-saving opportunities in a way they could understand and act on.
Up-to-date Cost Savings Information:
By leveraging Arcus’s automatic data discovery and continuous data integration, PumpGPT leverages the latest and most relevant domain-specific data available, giving customers continuous up-to-date, factual recommendations and preventing the model from going stale.
Achieving AWS Certification:
The Pump team set a crucial benchmark for success - passing the AWS Certified Solutions Architect Associate exam with PumpGPT, an exam that no-out-of-the-box language model could pass. Using Arcus, their model's performance surpassed expectations. The data provided and integrated via Arcus significantly improved the model's understanding of the domain, resulting in a 90% pass rate on the exam.
By harnessing the right data that the model actually needs, PumpGPT experienced significant improvements in its ability to respond with factual, personalized and actionable recommendations for Pump’s customers.
With Arcus, AI applications are grounded in factual real-world context, enabling them to produce high quality, accurate responses without hallucinating. The seamless integration of Arcus showcases how integrating the right data into your AI application can supercharge your AI efforts to drive significant performance improvements and empower your models to perform their best.
Interested in what Arcus can do for your AI applications?
Want to get in touch? Reach out at hello@arcus.co!
.svg)
