In case you missed it, I recently published a blog that quickly recapped our product progress in 2020 and then provided a brief glimpse into our product vision, strategy and key roadmap deliverables for the coming year. In that blog, I discussed four key areas of investment for Nuxeo, including:

  1. Artificial intelligence
  2. Low-code development
  3. Cloud operability
  4. And, connectivity

Beginning with this blog and over the next couple weeks, I will dive deeper into each of these critical topics and give you more insight into the future of our products and our roadmap. So, let’s start with one of my favorite topics, artificial intelligence and our Nuxeo Insight offering.

Insight: Nuxeo AI/ML Offering

As a quick refresher, Nuxeo Insight is our proprietary AI/ML offering - based on open standards like Tensorflow, of course - that we released in 2020 after beta testing in 2019. Nuxeo Insight is a cloud service that enables our customers to train custom Machine Learning models utilizing their own content and data. Fundamentally, this service is based on the premise that what we call “business specific” ML models are capable of delivering much better business value than commodity AI/ML services. This is because custom ML models can support each organization’s unique vocabulary and data requirements for its content or digital assets.

It is also important to note that we do support integration with a variety of different third-party cloud services for AI/ML, including offerings like Google Vision, Amazon Rekognition, Textract, Transcribe, and many others. And many of our customers use these services in combination with Nuxeo Insight.

A Unique Strategy for our AI Efforts

In 2020, we focused our roadmap for Nuxeo Insight on making this important cloud service accessible to business users. In particular, we introduced an entirely new, “point and click” interface that enables non-data scientists to quickly and easily configure, train, deploy and monitor the performance of custom ML models.

In the first part of this year, you will see us focus on more tooling for Nuxeo Insight that will support active learning and greater human interaction with the technology. In particular, we will add a number of enhancements for “human in the loop” processes that are intended to make it easier for humans to validate machine-generated values and to therefore continuously improve the accuracy of our Insight models. In the second half of the year, we will shift our focus to new platform capabilities enabled by AI/ML that will bring greater automation, efficiency and insight into working with content and digital assets.

Key Roadmap Elements

Following are the key roadmap items that we are actively working on or are planning for the next 12-18 months:

  • Human in the Loop. It is important to note that we support active learning and “human in the loop” validation today. However, this is through our standard Web interface, where predictions or suggested values are displayed alongside configured metadata fields and include a graphical indication of the degree of confidence in the prediction. Metadata values can also be automatically applied based on a configurable degree of confidence (i.e., we only ask users to validate data values where we have a lower degree of confidence).
    However, what we have realized is that there are more efficient ways to solicit “human in the loop” validations. The work we are doing now is to provide a dedicated UI/UX for user validation with a very simple “single query and yes, no, I don’t know” paradigm for confirming data values. For example, the query might be, “Is the talent in this picture Mary Smith?” and the user is simply given three possible responses. This highly efficient user interface is responsive, enabling users to perform validations on smart phones and other mobile devices. And, behind the scenes, there is an intelligent algorithm that is identifying and prioritizing content and assets for validation to further enhance the accuracy of the underlying ML model.
  • Gamification. To complement the work we are doing with “human in the loop” validation, we are also working to develop a net set of interfaces and dashboards that will provide rewards and recognition for users that are validating machine-generates values and are actively contributing to the long-term accuracy of our Insight models. These interfaces will enable administrators to define “missions” which are intended to produce specific business results and improvements in the Insight ML models. They will also provide a “gamified” experience for end users, recognizing those that are most actively contributing and also providing direct feedback on the impact their contributions are having. I have to admit that I was initially skeptical about the value that gamification would deliver from a business perspective, but the feedback from our customers has been unanimously positive. By the way, we plan to deliver our new gamified user interfaces in two phases over Q1/Q2’21, so watch this space.
  • Smart Governance. Shifting gears a bit, now let’s look at some of the AI-enabled capabilities that we are looking to incorporate into the Nuxeo Platform. We’ll begin with what we call “smart governance” capabilities. Specifically, Nuxeo Insight has proven to be very capable of auto-classifying both new and existing content and assets, and - once the object has been classified - in performing highly accurate metadata enrichment or entity extraction. In other words, Insight is very good at figuring out (automatically) what an item of content is and therefore how to accurately describe it with specific data.
    This is an outstanding foundation to drive all sorts of valuable downstream, governance activities. For example, we can utilize Insight to automatically identify vital corporate records and then apply the appropriate retention policies. We can also extend these capabilities for eDiscovery and applying legal holds. Similarly, we can utilize Insight to identify archival content and automatically move these objects to lower-cost storage tiers (particularly valuable for cloud customers). Now, it’s important to note that we can do all of this today with Insight and the Nuxeo Platform (and a bit of Services work), but the work we are doing here is to plumb these capabilities into the Nuxeo Platform and to provide a set of configuration tools that will enable records managers and other administrators to quickly configure our AI/ML capabilities to intelligently apply legal holds, retention rules, and archiving policies.
  • “Over the Shoulder” Training. Also referred to as “machine teaching,” this is a critical new capability for customers who work with a lot of different forms and frequently encounter new form types. Essentially, there are two ways of training a Machine Learning model: one, throw a lot of documents and data values at it and let it figure it out; or, two, through a guided human interaction, quickly demonstrate how to identify a form type and, perhaps more importantly, the specific data you want to extract from it. We refer to this second type of ML model training as “over the shoulder” training where a human operator can quickly and intuitively instruct a ML model to effectively process a new form type. In 2021, we will further explore these human-to-machine interactions and begin to develop an intuitive set of interfaces to enable over-the-shoulder training for Nuxeo Insight models.
  • Intelligent Knowledge Management. Another area of exploration for us is to begin to leverage Nuxeo Insight to map and understand the relationships between users, between users and content, and even between content objects themselves. We believe this is the vital foundation for AI-enabled knowledge management and accurate, predictive delivery of information to specific users. For example, if Insight recognizes the relationship between users, perhaps based on their role or observed behavior in the system, it can begin to make content recommendations based on what information users with similar profiles or interests consume (security permitting, of course). Similarly, if we can identify relationships between different content objects, we can also recommend additional content based on what a particular user consumes. And, if we begin to understand content in the context of a particular task or work activity, we can also automatically surface content to better inform decisions and accelerate task completion. The fundamental premise of Enterprise Content Management was always about delivering the right information, at the right time, to enable the right decision. With Nuxeo Insight and Content Services, we now have the ability to fulfill this simple, but powerful vision for content management.
  • Information Security. Lastly, and as an extension of the work we are doing to map user behaviors and content relationships, we also plan to explore more out-of-the-box features for AI-enabled content security. If, for example, Insight identifies an unusual pattern of user behavior (e.g., unusual downloads of large volumes of content, access to unusual types of content, attempted access to secure information, etc.), we want to give administrators a broad set of configurable options to address information security concerns. This could include temporarily disabling access to the system, flagging suspicious activities and launching investigative workflows, automatic alerting, as well as packaged dashboards and reports.

A Quick Summary

As you can see, we have ambitious plans for Nuxeo Insight and for how it interacts with our Nuxeo Platform. Again, our vision is to continue to enhance this offering as a powerful tool for the business and to make our AI/ML capabilities even more accessible to business users. We are also beginning to shift our focus from tooling to providing enhanced capabilities, in conjunction with the Nuxeo Platform, to provide more out-of-the-box, AI-enabled applications for this amazing technology. And please come back next week as I will provide more insight (sorry, pun intended) into our low-code roadmap for 2021 and beyond.

We’re open & transparent.