VentureBeat presents: AI Unleashed - An exclusive executive event for enterprise data leaders. Network and learn with industry peers. Learn More
Got It AI, a conversational AI firm helping enterprises, has unveiled Agent Copilot, an AI assistant for customer service and sales operations.
Agent Copilot combines generative AI with guard rails to empower agents to quickly and accurately access information from complex knowledge sources, leading to enhanced customer interactions and response times, said Peter Relan, chairman of Got It AI, in an interview with VentureBeat.
This solution utilizes Got It AI’s unique TruthChecker AI, designed to identify and prevent inaccuracies in responses generated by large language models (LLMs).
Agent Copilot offers businesses the flexibility to customize their AI experience by selecting from a range of LLMs, including ChatGPT, GPT-4, Llama-2, MosaicML, and Flan-UL2. The choice of LLM can be tailored to meet the specific knowledge base requirements and measured performance against the relevant documents.
An exclusive invite-only evening of insights and networking, designed for senior enterprise executives overseeing data stacks and strategies.
“You’re the pilot and the AI is the copilot,” said Relan. “The big categories are sales and marketing, banking and customer service.”
Also, Agent Copilot can be configured to integrate with various data sources, such as PDFs, web pages, documents, and presentations, enabling comprehensive support for multi-turn dialogs based on the compiled knowledge base. For organizations with data privacy concerns, the solution also supports on-premise installation of enterprise-specific fine-tuned LLMs.
“The risks associated with incorrect information are simply too high,” said Relan. “With Agent Copilot, we offer businesses the opportunity to choose the most optimal LLM for their needs, ensuring the highest level of accuracy and efficiency in their customer interactions.”
Relan also highlighted the impact of Got It AI’s TruthChecker, which significantly decreases hallucination rates in responses generated by the base model, resulting in near-human level accuracy for multi-turn dialogues in knowledge retrieval scenarios.
“The copilot can help them answer questions, like whether you have a deductible for your insurance or you can use out-of-network doctors or those things,” Relan said. “The agent typically has access to a knowledge database, or set of documents that could be web pages. Instead of searching through all that, suppose you can surface it in a conversational way. The copilot’s job is to go get the answer and come back with it.”
That makes the agent more productive, he said.
“We’re seeing beta customers deploy this because the alternative is longer calls,” Relan said.
I asked if the pricing was getting lower or was still under control. Relan said the costs are starting to move lower for such queries.
And since Relan’s firm was aimed at stopping generative AI from “hallucinating,” it is also mindful of the “guardrails,” Relan said. Those need to be in place to make the tech transparent so that if it does come up with a wrong answer, it won’t be like a “black box” where you can’t tell where it came from or if it’s wrong.
“We have something we call a truth checker, which will actually check the answers and provide an audit trail,” he said.
Agent Copilot is built on Got It AI’s recommended future-proof architecture, Enterprise Large Language Model Architecture (ELMAR), and is currently undergoing an open beta program with several enterprises for Agent Assist use cases.
The solution is expected to be widely available in September 2023. Its features include knowledge base connectors to custom data sources and documents, a choice of LLMs, TruthChecker AI for hallucination detection, fluid conversational dialogues with mitigation and disambiguation, active learning and fine-tuning for optimal accuracy, on-premises deployment with open-source LLMs, and seamless integration with other agent, CX, and CRM solutions.
Relan said there are large enterprises, financial services firms and insurance firms in beta testing.
“They are basically saying without the guardrails, we would not be considering this,” he said.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.