Digital Office Company

Solution

Digital Office Company collaborated with IBM technical teams on a six-week pilot that combined traditional machine learning and generative AI for document understanding. The implemented pipeline uses IBM Watson Discovery as the backbone of the data and retrieval flow, while IBM watsonx.ai and the Mixtral-8x7B large language model add generative reasoning on top of custom classifiers and metadata enrichment. The pilot focused on real-estate documents, where customers faced diverse formats and strict regulatory requirements. Documents enter the pipeline, are classified with custom models, enriched with labels and metadata, and then passed through generative AI to extract additional insight from content rather than relying on superficial tags alone. By combining deterministic classification with LLM-based interpretation, DOC improved both speed and document understanding while preserving the structure required for downstream compliance and business workflows.

Data flow

Incoming documents are classified with custom models in a Watson Discovery-backed pipeline, enriched with labels and metadata, and then processed with watsonx.ai and Mixtral-8x7B to extract deeper content understanding for downstream use.