Guild Takes Personalized Career Recommendations to a New Level with Arcee AI’s Custom Language Model

Guild was seeking an AI solution that would automatically provide consistent, customized career recommendations adapted to the diverse and dynamic backgrounds of each user.

Like many companies, they didn’t have enough expertise or bandwidth to build in-house. Another challenge: they are deeply committed to protecting their customer data, and found that traditional LLM providers like OpenAI and Anthropic could not guarantee data privacy and security.

Arcee AI’s small language models (SLMs) quickly emerged as the obvious solution:

• Unlike with traditional large language models (LLMs) offered by other AI providers, the SLM could be easily trained on Guild’s data, fine-tuned for this specific use case, and re-trained as needed at a low cost.
• The Arcee AI SLM provides full ownership and transparency, eliminating data security and privacy concerns.  ‍
• SLMs are extremely compute-efficient and cost-efficient, with a total cost of ownership (TCO) that’s significantly lower than that of LLMs. 

The results exceeded Guild’s expectations: the SLM successfully standardized career recommendations while customizing them based on individual metadata.

Guild also reported: 

• Increased user satisfaction thanks to the SLM's ability to provide highly-personalized career guidance.
• The SLM was evaluated against an existing closed source LLM+RAG solution–showing an “on par or better” score in more than 97% of responses.
• The Total Cost of Ownership (TCO) of the Arcee AI SLM architecture was estimated at a 47% reduction in costs when compared to on-demand costs of a closed-source LLM (such as OpenAI and Anthropic). There were also additional cost savings due to a reduced dependency on RAG infrastructure.

Guild deploys an Arcee SLM built on AWS to ensure they have the most secure, resilient, and cost-effective environment for their domain-specific Generative AI models.

With the Arcee SLM built on AWS, their data never leaves their VPC.