In Singapore today, just hosted our 4th CAST (Coffee And Showcase Tour) meeting "LLMs in Research - What Works and What Doesn't", following similar small-group sessions in San Francisco, London, and Hong Kong.

Within the domain of market research, the need for domain specificity, proprietary data, human expertise as a key factor in tweaking LLM deployments, and deterministic rules augmenting LLMs remain hot topics consistently across meetings.

Our use cases and experience at Robotic Online Intelligence (ROI) served as discussion triggers, as we grouped the cases into 3 categories:

1) Efficiently tapping into open-source intelligence (unstructured information);

2) Interrogating a specific knowledge base with RAG (Retrieval Augmented Generation);

3) Co-pilot-style interrogation of documents / human analysts working with larger documents (e.g. SEC filings, UK Companies House documents, PDFs of annual reports from SG REITs).

Some illustrations in the select slides: