For security and privacy, each instance of Odyssey AI includes its own privately hosted Language Model (e.g. Llama3). The language model used by default can be changed upon request at an additional cost by the owner of the Odyssey AI instance.
• Fine-tuning/Pre-train
We currently do not fine-tune an LLM. We use them as is. Fine-tuning an LLM is possible, this will result in a domain-specific private LLM(s) for Odyssey AI. But this requires compute resources and periodic re-training of the fine-tuning to ensure that it remains relevant to the changing domain-specific data. This is an expensive proposition for a private LLM.
For large enterprise customers this option can be offered for additional fees.
• Sync of new data within Odyssey AI
Odyssey AI implements Active Learning using topic models. The process of ingesting a document/data into Odyssey AI involves extracting the data from the data-source using our proprietary DocuTableAI algorithm, learn the topics using the Topic Model(s) and then using another proprietary algorithm InteliCompress to dynamically chunk the data and retain what is important and discard noise. This is done for every document or file that is ingested into Odyssey AI. So, when a new file or document is added to the data-sources integrated with Odyssey AI, the ingestion process is automatically trigged to synchronize the newly added data into the existing vectorized data within Odyssey AI.