• Amit Kukreja

Why Does Palantir's Ontology Matter?




Everything below in the article comes from Palantir's dedicated new documentation released on understanding their product better that can be found here.


I tried to organize parts of it so they make sense - I'll come out with more of these compilations on the blog since I know it may be difficult to dig though the actual website and find these resources.


From the website on why an ontology matters for a business, with use cases:


Organizations can gain several key benefits from building and using an Ontology to organize and leverage their data, as described below:

  • Connectivity at scale

  • Interpretability

  • Economies of scale

  • Decision capture

  • Powering operational AI/ML

In practice, these benefits are realized by using Foundry's Ontology-aware applications which enable rapid analysis, workflow development, and decision capture. Learn more about Ontology-aware applications.


Connectivity at scale

The Ontology is a shared source of truth for decision-making and decision capture across a large organization.


By providing a single source of truth, the Ontology enables business users to easily discover and understand the data available across their business as well as view their local decisions in a more global context, providing connectivity at scale. The Ontology is used not only to read data, but also to write data back and capture decisions made by users.


Operating from standard data lakes can lead to unmanageable complexity from an ever-growing number of datasets, dashboards, and applications. Over time, increasing effort is required simply to understand what data assets exist or should be used, while new projects "reinvent the wheel" instead of reusing or leveraging existing data assets.


In contrast, the Ontology provides a well-defined system into which new information is modeled into a common language for the organization. With an Ontology, organizations can make the most of their data as the data asset grows, enabling a digital transformation at scale, while controlling complexity and reducing the difficulty of data management.


Example:


An energy company uses the Ontology to create a shared view of the health and performance of a well across petroleum engineers, well integrity engineers, and well management staff. Instead of building several isolated views on the well’s performance, they share their inputs to the same Ontology well object type, allowing short-term decisions about the well’s management and long-term decisions around the asset’s investment strategy to be generated from the same information and insights.




Interpretability


The most challenging element of operating as a data-driven organization is deploying the data to the various decision makers across the organization. In particular, many decision makers are not technical users comfortable with code or IT concepts such as datasets or joins.


The Ontology abstracts away these digital concepts and allows business users to engage with data represented in the standard business terms they use every day. More importantly, the Ontology provides a shared language across different users and functions, allowing them to collaborate without lengthy reconciliation processes to confirm that everyone is looking at the same information.


Example:


At a manufacturing customer, monitoring data from aircraft sensors was previously inaccessible to both non-technical users and data scientists due to its scale, complexity, and esoteric format. Today, because of the Ontology modeled on top of the monitoring data, designers can search for parts, see related sensor readings that may signal unexpected or anomalous behavior, and improve future designs without needing to think about tables or joins or going through a complex data preparation process. Although these data were among the most valuable and useful in the organization, under the previous system the process of preparing the data could take days or weeks, limiting use of the data to special projects. Now, these data are immediately accessible not only to data scientists, but to engineers, quality control specialists, and designers.





Economies of scale


The Ontology enables significant economies of scale in the construction of an operational platform by converging effort onto a single reusable data asset that supports all analytical work and application development.


Rather than requiring a dedicated data integration and data layer effort for every new use case or project, data integration is only required for new data entering the platform. Entire applications and use cases can be built on the existing Ontology; the shared data asset lets application builders focus on the business problem and the user workflow, instead of on data wrangling.


Decision capture


As an organization's "digital twin", the Ontology supports data writeback and continuous improvement by capturing decisions being made in the organization as data. The Ontology allows for the configuration of writeback and action types, which define how users can edit and enrich the data backing the Ontology.


Capturing decision-making outcomes in the Ontology enables organizations to learn from and improve their decision making. Data writeback also allows the value of the data asset to compound over time, as insights captured by one user can contribute to the decision-making of another user.


Powering operational AI/ML


For data science and AI/ML teams, the Ontology enables collaboration with business and operational teams on a shared platform. Models (and their features) can be bound directly to the building blocks and processes that drive the business. This allows models to be governed, released, and implemented directly into core applications and systems, without additional adapters or glue code, before being served in-platform (batch, streaming, or query-driven) or externally. As decisions are made and actions are taken, operational and process data are written back into the Ontology, creating a feedback loop that enables model monitoring, evaluation, re-training, and MLOps.



Foundry enables quick iteration towards outcomes. The Ontology and other best-in-class tooling make it easy to get started and deliver AI/ML-enabled operational outcomes, whether through new applications or by augmenting existing systems. Subsequent use cases can leverage interconnected datasets and model assets throughout the enterprise, decreasing time-to-value for new projects.



Thanks for reading the article. If you'd like to get in contact, please @ me on twitter here or email me at amit@dailypalantir.com.

226 views