InfoMind AI Enterprise: Giving Unstructured Information an Intelligent Mind

InfoMind AI Enterprise aims to transform unstructured organisational documents into searchable, organised knowledge. The system employs local LLMs, hybrid search, and grounded RAG, a technique intended to ensure the AI assistant's responses are based directly on provided documents. The developer reports the system is designed to state when insufficient evidence exists rather than fabricating answers.

Internal testing with the prototype suggests the choice of LLM significantly impacts its ability to answer questions accurately and avoid erroneous refusals. An initial 'small default model' reportedly failed to answer a specific question in five attempts when less relevant information was also retrieved. However, switching to Llama 3.1 (8B) enabled the system to answer the same question successfully in ten out of ten tries.

Further internal evaluation with the Llama 3.1 (8B) model indicated strong reported performance in distinguishing between answerable and unanswerable questions. The system reportedly answered four out of four questions for which evidence was available and correctly refused three out of three questions where information was lacking. These results are based on synthetic data and made-up office files, not real-world organisational documents.