Reimagining Startup and Enterprise Innovation

Deploying Bespoke AI using Fn Project - KADlytics by Miminal

Guest Author

This blog post was authored by Will Worrall, Director & Data Scientist at Miminal


At Miminal, one of our largest projects is the commercialization of bespoke artificial intelligence which reveals hidden dependencies in complex engineering projects. The product that does this is KADlytics, and more can be found out about the product at www.kadlytics.com.

So, how is this done? The AI is built on the principle that if two assets (e.g. CAD model files, documents etc.) are updated at approximately the same time then this is a proxy for some kind of dependency. For example, employee A makes an update to their CAD model and notifies employee B that they must update their CAD model, to account for A’s changes. Employee B does this at the next possible opportunity. Whilst this dependency relationship is simple, organizations suffer from not understanding these dependencies in enough detail in large collaborative projects. This often leads to unforeseen work having to be carried out after an update to an asset, to update unforeseen dependencies.

KADlyitcs learns and tracks a project dependency network using just update metadata as input, namely the timestamps of asset updates. It then puts this network to use, alerting employees of whose work they may affect when they change an asset, or if their work must be updated due to an asset change. Beyond digital assets, KADlytics also reveals the dependencies between people and organizational departments. In addition, it aids the project management team in predicting the project-wide design effort (cost) created by an update to a particular asset in the project.

In order for KADlytics to operate on a project, several stages of processing must occur:

  1. On project initialization, all the project's asset’s update metadata must be fetched from the cloud storage API. Currently, KADlytics is integrated with Autodesk cloud storage.
  2. Next, the bespoke AI algorithm operates on the database of update metadata, outputting the project’s dependency network.
  3. New updates must be streamed to KADlytics in real-time. This is achieved by registering webhooks with Autodesk, and hosting an endpoint to receive and act on them.
  4. Stage 2 Must be repeated periodically to reflect project updates.
  5. Network theory operations and queries must be carried out on the project dependency network to give meaningful information to KADlytics users.

Each of these stages have 3 things in common:

  1. Processing is batched, varying from short-lived to medium-lived processing.
  2. Computational workload is unique.
  3. Runtime environment and set of software libraries is unique.

These commonalities form a set of requirements for the hosting infrastructure. Our research showed that Fn Project is the only solution that meets these requirements, scales, doesn’t lead to unused compute capacity at times of low traffic, is failure resilient and supports any desired runtime environment with its revolutionary use of Docker containers as serverless functions. In addition to this, using Oracle Cloud Infrastructure to host our Fn Project servers means our serverless functions run on physical servers with an underlying compute-power far greater than the commodity servers of other cloud providers.

Finally, Fn Project has made developing a flexible, modular and extensible microservice architecture a pleasure. Go and check it out at https://fnproject.io .


Be the first to comment

Comments ( 0 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.