Gretel, a provider of synthetic data generation, is releasing Gretel Navigator, an agent-based, compound generative AI system built to automate data creation and curation processes for AI development.
With simple natural language or SQL prompts, Gretel Navigator enables users to create, edit, and augment tabular data, and design realistic, high-quality test and training datasets from scratch, according to the company. Developers can also leverage existing datasets to generate insight-rich synthetic data on demand.
"Whether you're building a retrieval-augmented generation (RAG) system, training a foundation model, or fine-tuning an LLM for a specific task, high-quality data is the single most important ingredient for success," said Ali Golshan, co-founder and CEO at Gretel. "But the status quo for acquiring that data is broken. Scraping the web leads to inconsistent quality, and de-identifying private data does not offer adequate protections. Meanwhile, manual data labeling is time-consuming and costly, translating into weeks and months of data preparation before the real work can even begin."
Gretel Navigator addresses traditional challenges with data acquisition head-on by enabling developers to generate customizable, realistic synthetic datasets that mimic real-world patterns without compromising individual privacy, the company said.
Navigator supports a wide range of data formats, modalities, and context-specific optimizations to streamline workflows and expedite AI projects.
Gretel Navigator is powered by an ensemble of pre-trained AI models, including Gretel’s custom tabular Large Language Model (LLM) which was trained on a diverse curation of public and proprietary datasets, including electronic health records, financial documents and market data, and other industry-specific formats. This enables the system to generate high-quality, vertical-specific, synthetic tabular data that is crucial for enterprise AI applications, according to the company.
Gretel Navigator incorporates privacy-enhancing technologies, such as differential privacy, and addresses key AI development challenges, such as domain knowledge gaps and historical biases in limited real-world datasets. It also prevents issues like model drift and boosts overall model accuracy for high-value AI applications.
By enabling secure, real-time access and tailored optimizations of sensitive or proprietary training data, Navigator empowers developers to build state-of-the-art models that are continuously learning and adapting to critical real-world feedback, according to the company.
For more information about this news, visit https://gretel.ai/navigator.