How We Use AI: Open-Source Models & Roadmap
At Zoho, we have been investing heavily in Zia for more than 13 years to provide contextual intelligence across our suite of business applications. Our priority has always been to build deep technical know-how in-house and stay abreast with the latest technological advancements to maximize value for our customers.
Our AI strategy for LLMs also follows this core principle of Zoho. While we are actively investing in a long-term AI strategy to build proprietary large language models, we are leveraging open-source models to meet the immediate needs of our customers.
After careful evaluation, we have selected Llama-3.1 as our primary model owing to its flexibility and precision. We have invested heavily in rigorous testing, continuous evaluation, and use-case-specific fine-tuning to meet enterprise standards. We take prompting as our primary approach, as it effectively addresses a wide range of use cases. However, in scenarios where prompting alone is insufficient, we incorporate LoRA-based fine-tuning to improve accuracy and relevance.
Additionally, as new AI models emerge, we will continue to evaluate and refine our approach to deliver value from new AI developments to our customers.
Llama-3.1 is a large language model developed by Meta and released under the "Llama 3.1 Community License Agreement". Below, we list the AI features across different Zoho applications that are "Built with Llama version 3.1".
| S.No | Products | Product Use cases |
|---|---|---|
 | Zia helps provide a chat summary feature where users quickly catch up on ongoing conversations in a chat or channel.s | |
 | Zia helps summarize long posts to make it easy to read the core content of the post. | |
 | Zia streamlines application development by understanding natural language inputs, generating relevant use cases, forms, and fields, while enabling effortless form creation with intuitive prompts and intelligent field suggestions for a seamless experience Zia helps in application development and form creation. | |
  | Zia identifies user intent and triggers the appropriate actions within the chatbot for a seamless interactive experience. Zia provides a quick summary of ticket conversions, enabling users to review key details at a glance. | |
 | Zia makes form creation intuitive, fast, and efficient. Simply describe your form in plainlanguage, and Zia will generate a customized form to match your needs. Zia helps generate well-crafted email content effortlessly. | |
 | Zia leverages AI to generate use cases from user provided descriptions and build a base model tailored to those scenarios. | |
 | Zia is used for note summarization, content generation and note completion. | |
 | Zia generates content based on a given description and provides the option to store the generated content as a file in WorkDrive. AI-powered comment suggestions improve team collaboration by refining comments in real time. | |
 | Zia analyses customer tone, provides reply assistance, and enhances writing for ticket thread content. It also supports the RAG Answer Bot for accurate and context-aware responses. Zia generates summaries for ticket conversations and thread content to provide a quick overview of key details. | |
 | Zia is used for generating content on a given topic and responding to user prompts. Zia is used for punctuation, rephrasing, shortening, suggesting headlines, listing keywords, and recommending single-word alternatives for selected content. Zia generates a concise summary for selected content or the entire document. | |
 | Zia generates business descriptions and replies for reviews, offering quick and relevant responses. Zia rephrases business descriptions and review replies to ensure clarity and improved tone. |