Bringing the Value of Self-Hosted LLMs to Financial Services

Why industry models and fine-tuning is not the answer

As the landscape of alternative asset management continues to evolve, the intersection of technology and data stands as a cornerstone for innovation and maintaining a competitive edge. Many financial services are looking to external and internal data sources to capture trends, uncover information, and provide insight around operational opportunities. Large language models (LLMs) serve a growing use case around this data, as they provide an amazing ability to summarize unstructured data.

While existing models like BloombergGPT have demonstrated remarkable capabilities, they often fall short in addressing the nuances of organizational-specific use cases for LLMs. Fine-tuning improves model performance, but has its limits due to the specific training the fine-tuning is specialized for. Enter Retrieval Augmented Generation (RAG).

RAG fundamentally improves use case performance by enabling LLMs to access organizational context in real time. Whether it’s industry news, market data, recorded earnings calls, or internal emails, these data sources can provide needed context for the LLM. The Shakudo platform is a catalyst in the realization of RAG-based LLM architecture for financial services .