Gretel.ai | Making Data Work

TL;DR

Gretel.ai makes data safer to use by helping organizations evaluate quality, identify viable use cases, apply privacy controls, and create synthetic alternatives to restricted or incomplete datasets. Its first public beta arrived in September 2020, nine months after the founders began writing code. The platform targets the access and tooling gap between large organizations and smaller teams, making the details of its synthetic-data approach worth examining.

Transcript

hi everyone welcome to gray matter the podcast from greylock where we share stories from company builders and business leaders i'm heather mack head of editorial at greylock today i speak with ali gulshan who is a ceo and co-founder of gretel ai similar to the way github's open platform unlocked coding and computing bottlenecks for engineers gretel... Read More

Key Insights

• Code and data differ: GitHub demonstrated how shared repositories and collaboration could democratize building with code. Gretel adopts that platform ambition for data, but Golshan emphasizes that information has higher entropy and is constrained by ethics, privacy, regulation, compliance, and company policy. Those constraints make removing the data bottleneck a distinct problem.
• Access shapes experimentation: The founders had worked in small companies where inadequate data prevented testing and validation, then inside larger organizations with abundant information for research and development. That contrast showed them that data access affects whether teams can evaluate ideas quickly, not merely whether they can train a final machine-learning or artificial-intelligence system.
• Advantages can compound: Organizations with superior tools can improve their data, and improved data can support even stronger tools. Golshan describes this as a self-fulfilling cycle that favors larger companies. Gretel’s response is to make sophisticated capabilities available beyond the protected environments where extensive data and specialized infrastructure already exist.
• Community data has collective value: Although individual companies can build walled gardens, Golshan argues that the broader community collectively has more data than any single organization. Unlocking that value requires tools that support safe sharing and collaboration. Gretel therefore focuses on the infrastructure around data use, not only on producing synthetic records.
• Expertise requirements restrict adoption: Differential privacy and complete data pipelines can demand expertise that many teams do not possess. Gretel’s founders wanted these capabilities to be usable without multiple PhDs or substantial custom infrastructure. Accessibility is therefore part of the company’s proposed solution to the data bottleneck, alongside privacy, quality, and sharing controls.
• Three questions define usefulness: Gretel’s early users first wanted to know the quality of their data, then which use cases that quality could power, and finally how private the information was. These questions connect technical measurements to practical decisions. A dataset is not useful merely because it exists, since teams must know what they can safely do with it.
• Privacy determines sharing scope: Gretel treats privacy as a question of permissible audience. Users need to decide whether data can remain within one team, circulate throughout a company, or be published on the web for wider collaboration. Quantified privacy characteristics help make those boundaries visible instead of leaving sharing decisions to unsupported assumptions.
• Synthetic data expands access: When permissions limit the original dataset to a small group, a generated alternative can support wider experimentation and collaboration. The value depends on users understanding its quality, suitable applications, and privacy properties. Gretel therefore presents synthetic generation as part of a broader evaluation process rather than as an isolated replacement step.
• Representation can be strengthened: Synthetic generation can add examples to underrepresented portions of a dataset. This matters when limited relevant records could produce biased answers or decisions in machine-learning and artificial-intelligence applications. The approach gives teams another option besides assembling specialized staff and extensive infrastructure to repair representation gaps.
• Testing can precede collection: Organizations can generate test data before paying to collect, acquire, or sanitize the corresponding real-world information. This allows teams to explore whether an application is promising before committing to costly infrastructure. In Gretel’s framing, synthetic data can reduce the economic barrier to early validation while preserving attention to quality and privacy.
• Transformations can improve utility: Raw information may contain missing fields, weak classifications, inadequate labels, or uneven representation. Gretel’s approach can apply improved labeling and transformations before producing a synthetic dataset. The resulting data may be more useful than the untreated source when those steps directly address defects that would otherwise limit the intended use case.
• Transparency supports evaluation: Users told Gretel that judging the quality of artificial-intelligence tools and algorithms was difficult. The company responded by open-sourcing its core synthetic-data libraries and some natural-language-processing work. Community inspection allows claims about quality and behavior to be examined beyond Gretel’s own internal assessment, supporting the trust its platform requires.

Questions & Answers

Q: How does Gretel.ai make data safer to use?

Gretel.ai helps teams examine data quality, determine which use cases the data can support, and evaluate how privately it can be shared. Its privacy question is practical: whether information is appropriate for a team, an entire company, or publication on the web. Synthetic data can provide an alternative when access to the source is restricted, while labeling and transformations can address weaknesses in the original. These controls matter because data is governed by privacy, ethics, regulation, compliance obligations, and internal policies.

Q: Why is Gretel.ai called the GitHub of data?

The comparison reflects Gretel’s goal of creating an accessible ecosystem for collaboration around data. GitHub made repositories and cooperative code development easier, helping democratize the process of building applications and services. Gretel seeks a similar reduction in friction for data experiments and application development. The analogy has limits because data carries privacy, ethical, regulatory, compliance, and internal-policy constraints that make sharing more complicated than code collaboration.

Q: What problem does Gretel.ai solve for organizations?

Gretel.ai addresses the bottleneck that prevents teams from safely accessing, sharing, testing, and building with data. Smaller organizations may lack sufficient information, validation resources, specialized researchers, or the advanced tooling available inside larger companies. Gretel packages capabilities such as synthetic generation, differential privacy, quality assessment, and data pipelines so they are easier to use. The goal is to let organizations of different sizes experiment without first recreating the infrastructure of a large company.

Q: What are Gretel.ai’s three main synthetic-data use cases?

The first use case is expanding access when permissions restrict the original dataset to only part of an organization. The second is generating more examples for underrepresented portions that could otherwise contribute to biased results. The third is creating test data before investing in collection, acquisition, or sanitization infrastructure. Together, these uses address access constraints, representation problems, and the cost of validating an idea with real-world information.

Q: How can synthetic data improve an imperfect dataset?

A source dataset may have missing columns or fields, insufficient labels, weak classifications, or underrepresented segments. Gretel’s approach can apply better labeling and transformations while generating a synthetic version. Additional generated examples can strengthen relevant portions that are too small for reliable machine-learning or artificial-intelligence work. The result can be more useful than untreated raw data when the generation process directly addresses the source dataset’s documented limitations.

Q: How does Gretel.ai help teams address data bias?

Gretel can generate additional records for parts of a dataset that lack sufficient representation. Without that reinforcement, a model trained on limited relevant examples may produce biased answers or decisions. Synthetic generation lets teams strengthen those portions without first building a large specialist group or extensive internal tooling. The approach still requires teams to assess quality and confirm which use cases the generated information can legitimately support.

Q: Why did Gretel.ai open-source its core libraries?

Early users found it difficult to determine the quality of artificial-intelligence algorithms and tools. Gretel opened its core synthetic-data libraries and some natural-language-processing work so the community could inspect and validate them. This transparency provides a basis for evaluating behavior beyond the company’s own claims. It also supports Gretel’s broader promise to make data quality, suitable use cases, and privacy characteristics understandable to users.

Q: How did Gretel.ai’s early development shape the product?

Golshan and his co-founders wrote the initial lines of code in late 2019. They released the first public beta in September 2020, nine months after starting, and spent substantial time speaking with users and the community. Those conversations repeatedly surfaced questions about quality, usable applications, and privacy. Gretel used those questions to organize its platform, roadmap, open-source work, and approach to helping customers evaluate synthetic data.

Summary

In this video, Ali Gulshan, CEO and co-founder of Gretel AI, discusses the company's mission to remove the data bottleneck and provide fast and easy access to safe data through synthetic data. He explains the inspiration behind Gretel, the early conversations with investors and users, and the advantages of using synthetic data. He also touches on the use cases in healthcare, web 3.0, and other industries, as well as the future focus areas for the company.

Questions & Answers

Q: What does it mean when people say Gretel is the GitHub of data?

Gretel being called the GitHub of data refers to the company's aim to unlock and remove the bottleneck for data, similar to how GitHub unlocked coding and computing bottlenecks for engineers. Like GitHub, Gretel provides tools that allow organizations of any size to run experiments, collaborate, and build with data, democratizing the process of working with data.

Q: How did the founders of Gretel identify the data bottleneck issue and decide to work on it?

Ali Gulshan and his co-founders, Alex Watson and John Myers, came from the intelligence community and had firsthand experience of the challenges faced in working with limited access to data. They also experienced the advantages of having abundant data in larger companies. They noticed that larger companies were building walled gardens around their data, creating a major advantage for themselves. With the increasing importance of data in various industries, they saw the need to remove the data bottleneck and make advanced data tools accessible to everyone.

Q: Who were some of the early investors and users that helped Gretel build its product roadmap?

Sridhar Ramasamy from Greylock Partners joined Gretel as an investor during the seed round and played a key role in the early conversations. The conversations with early investors and users revolved around understanding the main questions users and customers needed to answer in order to remove the data bottleneck. Through these conversations, Gretel identified the need for tools that assess data quality, enable broader access to data, and ensure privacy. The insights gained from early users and customers helped shape the product roadmap.

Q: What are some advantages of using synthetic data?

Synthetic data can often yield better results than raw data, especially in cases where the raw data is missing columns, improperly classified, or labeled. Synthetic data goes through proper labeling and transformations, resulting in higher quality and better privacy. The tooling approach of Gretel, with better labeling and transforms, contributes to the superiority of synthetic data. By focusing on data quality, privacy, and use cases, Gretel aims to showcase the advantages of synthetic data and change the perception that raw data always yields better results.

Q: How does Gretel plan to make data engineering accessible to a broader set of users?

Gretel aims to make data engineering easy and safe to use for everyone, including individuals and smaller companies that may not have the resources to invest in specialized data teams. The platform provides a low-code, no-code approach to data engineering for tasks such as synthetics, labeling, and transforms. Users can sign up on the website, drag and drop their data files, and leverage the toolkit provided by Gretel. The focus on usability and user experience allows users to work with data without requiring extensive expertise.

Q: Are there any applications of Gretel in the field of healthcare?

Yes, healthcare and health sciences are areas where Gretel has found significant use cases. During the COVID-19 pandemic, the removal of red tape allowed health sciences, pharma, and hospitals to recognize the benefits of faster work with data. Gretel has been working with companies like Illumina, the world's largest genomics company, to build synthetic data sets for genotype and phenotype analysis. They have also collaborated with the University of California, Irvine on improving the detection of female heart disease. By combining synthetic data with other forms of data, Gretel enables more accurate predictions and research in healthcare.

Q: How does Gretel enable collaboration and data sharing in the healthcare field?

Gretel's platform allows users to generate synthetic data that can be shared and collaborated on, while ensuring privacy and providing a quantified view of differential privacy. This becomes especially valuable in healthcare, where regulations and privacy concerns are paramount. For example, researchers working on skin cancer can generate variations of skin cancer images and combine them with medical data, allowing for better context and decision-making. Synthetic data helps bridge the gap between different data types and enables collaboration among healthcare professionals and researchers.

Q: Are there any applications of Gretel in the web 3.0 space?

Gretel has started to see an uptake in web 3.0 companies, particularly in the areas of financial services and web3 gaming. These companies often face challenges with limited or sparse data and require better forecasting or testing capabilities. Synthetic data can help them generate high-quality data for better predictions. Gretel has been working with web3 gaming companies to provide synthetic data for their test networks, where the data needs to mimic the production data on the public blockchain. The unique data needs of web 3.0 companies create interesting use cases for synthetic data.

Q: What are the future focus areas for Gretel?

Gretel is focused on becoming a single platform for all types of synthetic data, allowing users to combine different data types, such as tabular, image, and visual data, for accurate predictions and training of ML and AI systems. They aim to provide deep visibility and usability for synthetic data, enabling users to make data operational quickly and easily. Additionally, Gretel is investing in usability and integration, aiming to become an easy-to-use tool that seamlessly integrates with other services and tools, such as storage systems, data lakes, and orchestration platforms. The goal is to automate complexity and make the platform a cohesive part of the larger data ecosystem.

Q: Who is Gretel looking to hire currently?

Gretel is rapidly expanding and looking to hire across various departments. They are particularly focused on hiring in engineering and applied research. The applied research team plays a vital role in the company, ensuring the development of advanced data tools and techniques. Gretel also places importance on building a people-friendly organization, and they are investing in their talent team to support this goal. As Gretel continues to grow, they aim to prioritize usability, integration, and scalability in their hires to meet the needs of their expanding user base.

Takeaways

Gretel's mission is to remove the data bottleneck and provide fast and easy access to safe data through synthetic data. Their platform, often referred to as the GitHub of data, democratizes the process of working with data and enables collaboration and sharing. Synthetic data offers advantages over raw data in terms of quality and privacy, and Gretel aims to educate the market about its potential. The company has seen success in healthcare, web 3.0, and other industries, and plans to become a single platform for all types of synthetic data. They are rapidly expanding and hiring across departments, with a focus on engineering and applied research. The goal is to make data engineering accessible to a broader set of users and build a cohesive tool that integrates with other services.

Summary & Key Takeaways

• Defining the data bottleneck: Grey Matter host Heather Mack introduces Gretel.ai CEO and co-founder Ali Golshan and frames the company as an open platform for building with data. Golshan says GitHub helped remove computing bottlenecks by making code collaboration easier. Gretel applies that idea to data, but data is harder to unlock because privacy, ethics, regulations, compliance requirements, and internal policies restrict how organizations can access, share, and reuse it.

• Tracing the founders’ experience: Golshan, chief product officer Alex Watson, and chief technology officer John Myers had known one another for 10 to 15 years. All three came from the intelligence community, founded companies, and later worked inside larger organizations after those companies were acquired. They experienced both limited access to data and the advantages of abundant information, strong testing resources, effective validation, and tooling that remained inaccessible to much of the broader community.

• Challenging protected data advantages: The founders observed a self-reinforcing cycle in which better tools produce better data, while more data enables organizations to build still better tools. Large companies consequently create walled gardens around their information and gain a substantial advantage. Gretel was conceived to give the wider community comparable tools for sharing, experimentation, and collaboration as sensors, IoT systems, and other collection methods generate additional types of data.

• Making advanced tooling accessible: Gretel set out to package capabilities that large organizations could build internally and make them easier for organizations of any size to use. Golshan specifically argues that teams should not need multiple PhDs to use differential privacy or construct complete data pipelines. The platform combines synthetic data, labeling, transformations, privacy controls, and quality assessment so teams can understand generated data rather than merely receiving an unexplained output.

• Building through user questions: The founders wrote Gretel’s initial code in late 2019 and released the first public beta in September 2020, nine months after starting. Early users repeatedly asked how good their data was, which use cases its quality could support, and how privately it could be shared. Gretel organized its work around those questions and open-sourced core synthetic-data libraries and some natural-language-processing work so its algorithms and tools could be inspected and validated.