SentiaSys is a new AI company with an exciting mission: to move today’s blossoming AI capabilities from those data centers in the cloud down to everyday consumers. We are providing the capabilities that people want from AI on their mid-level computers without requiring special hardware or cloud subscriptions. As such, we are targeting high-volume applications that are highly desired but currently held back by hardware limitations.

We are diving into the core reasons why small Large Language Models (LLMs) have not really entered the consumer market — chief of which is that they are not sufficiently prompt or reliable for those applications. There are many reasons why this is so, but the primary one is the tendency for hallucinations in these models. This can make long interactions with models go significantly off the rails. This undesired behavior tears users out of the useful and immersive experience that is intended. So far, the response of the AI community has been largely one of moving the most recent (and largest) model. But there are other approaches to conquer these excursions into non-factual territory. Here are some of our current efforts:

• Tiny Model: We have selected a highly performant small LLM which integrates well with external applications, with or without an available GPU, but is responsive enough for even gaming situations. Our target markets include products sold to people with standard PCs.
• Model Distillation: Get the desired behavior out of a larger model, then use distillation to teach a smaller model. This is enhanced with our own triply coupled distillation process which looks at relationships in embedding space, in graph space, and in probability space.
• World Knowledge Graph: Create a Knowlege Graph with a compendium of facts and relationships about the world that the LLM is to operate within. This is compiled with minimum effort by the customer, who needs only provide a sufficiently rich description of the entities and relationships between all elements of the product’s application.
• Retrieval-Augmented Generation: Use the world knowledge to fortify the prompts given the LLM with background information as well as material from prior discussions. We do this in a highly nested way within the LLM reasoning process itself, scoring useful responses much higher than non-factual responses, while still keeping the
• Human-Curated Experience: We use human quality control expertise to fine-tune our products and verify their quality. We also use skilled authors and artists to add creative and performative touches to the AI without disrupting the basic requirements of operating within a fact-based world which conforms to the application.
• Optional Voice and Visual Avatar Products: We enhance the unique personalities of our online entities with matching expressive and properly accented voices and highly detailed visual avatars which supply a wide range of visual expressions. While this increases the computational requirements of the system, it supplies an order of magnitude increase in the perceived personalization of all the entities, and creates vivid and memorable characters.

SentiaSys Enables Great AI Enhancements to our cusTOMER’s Products WITHOUT the Expense of DEVELOPING a DEDICATED AI TEAM

The name SentiaSys invokes the process of bringing Sentience to Systems. We are bringing AI to the systems used by ordinary people — to enrich their lives and even change our world.

We sincerely hope that SentiaSys’s focus on enabling AI technologies and packaging them in a convenient and powerful way will give our customers to focus on creating their own great products, for which we provide the tools to generate AI enhancements.