Got It AI ELMAR™
(Enterprise Large language Model ARchitecture)
Enterprise-grade Architecture to
Build ChatGPT-like Solutions
with Guardrails using Any LLM
Key Features and Benefits
1
On-Prem or Cloud
ELMAR can be fine tuned for tasks to produce task-specific custom models. These models can then be deployed on commodity hardware within your enterprise's security envelope.
2
Guardrails
Large Language Models often hallucinate and thus cannot be trusted for mission critical applications. ELMAR surrounds core LLM model with pre and post processor guardrails to minimize the impact of LLM unpredictability.
3
Choose The Optimal LLM
Choose between OpenAI, Meta, Google, Anthropic, Open Source or ELMAR fine tuned LLMs. We benchmark all leading LLMs to make sure ELMAR Guardrails work across all of them.
Architecture
ELMAR, short for Enterprise Language Model Architecture, is an enterprise-ready Language Model that enables businesses to create sophisticated chatbots that can communicate with customers or internal stakeholders in a natural and engaging way. You can choose to use models like GPT-3, GPT-4, Claude, Google PaLM through third party APIs, or use ELMAR fine tuned models for a cost-effective solution.
One of the most significant benefits of ELMAR is its ability to run on-premise using commodity hardware, which means that sensitive data need not leave the company's premises or VPC. This provides businesses with a sense of control and safety around the use of generative AI models and ensures data privacy and security.
All LLMs are known to make up facts or "hallucinate". We benchmark third party and our own LLMs for hallucination rate, i.e. what fraction of responses contain inaccuracies.
One of the guardrails ELMAR brings is TruthChecker. It detects hallucinations in LLM generated text and can thus lead to 50-80% reduction in hallucinations. This feature helps to enhance the reliability of chatbot interactions with customers or internal stakeholders and improve overall customer experience. For OpenAI LLMs, ELMAR’s hallucination guardrails reduce hallucinations down to human level performance, as benchmarked on hundreds of articles of real customer data.
* Hallucination rate measured on a set of 530 Q&A pairs over a 200 document knowledge base from a Got It AI customer
For Open Source LLMs, ELMAR comes with a choice of models such as Llama2 and MPT, as well as a native 3B parameter LLM which delivers the lowest hallucination rates among on-prem model options, while being substantially smaller.
* Hallucination rate measured on a set of 530 Q&A pairs over a 200 document knowledge base from a Got It AI customer
Overall, ELMAR offers a unique value proposition for businesses looking to leverage the power of AI and chatbots while maintaining data privacy and security. With its ability to run on commodity hardware and TruthChecker feature, ELMAR is a cost-effective and reliable solution that can help organizations streamline their operations and drive growth.
Integration with NVIDIA AI Solutions
AS FEATURED ON
