aiSSEMBLE is Booz Allen's lean manufacturing approach for holistically designing, developing and fielding AI solutions across the engineering lifecycle from data processing to model building, tuning, and training to secure operational deployment. This repository consists of standardized components which make it easy for dev teams to quickly reuse and apply to their project to drive consistency, reliability and low delivery risk. aiSSEMBLE offers projects the rapid generation of necessary scaffolding, boilerplate libraries, and container images with the flexibility of custom configuration. It consists of pre-fabricated components that can be used as is within your projects and generated capabilities that can be extended.
Many languages can be useful across the full breadth of AI solutions. Currently, the following languages are leveraged:
- Data Delivery / Machine Learning Inference
- Java
- Python
- Machine Learning Training
- Python
In addition, the following build tools and container frameworks are an important part of aiSSEMBLE:
- Fermenter MDA
- Maven
- Habushu Maven Plugin (builds Python modules)
- Orphedomos Maven Plugin (build Docker modules)
- Helm Maven Plugin
- Kubernetes
- Helm
aiSSEMBLE documentation is available GitHub pages.
aiSSEMBLE is currently released about once a month, but we intend to increase to around twice a month as we get our processes adjusted and honed into the public GitHub
Please consult our Configuring Your Environment guidance.
The following steps will build aiSSEMBLE. You must follow the configuration guidance above first.
- To get started, pull the latest code for the aiSSEMBLE repo from git.
- Configure ghcr.io authentication SNAPSHOT repository support - server configuration is all you need, you can ignore setting up a repository
- Ensure Rancher Desktop is running.
- Build the project locally using the
./mvnw clean install
command.- A successful build will have an output similar to the below.
[INFO] ------------------------------------------------------------------------ [INFO] BUILD SUCCESS [INFO] ------------------------------------------------------------------------ [INFO] Total time: 10:16 min [INFO] Finished at: 2021-09-09T10:01:10-04:00 [INFO] ------------------------------------------------------------------------
The aiSSEMBLE baseline project provides several build profiles that may be helpful for different development environments.
To activate each one, use the standard Maven syntax: ./mvnw clean install -P[profile_name]
, for
instance, ./mvnw clean install -PnoRdAdmin
. There are many profiles you can find in the root pom.xml
file. The
following profiles are often useful when first starting with aiSSEMBLE:
-
noRdAdmin: For configurations that disallow granting administrator privileges to Rancher Desktop. Testing frameworks
leveraged by aiSSEMBLE may, at times, assume that the docker unix socket is located at
/var/run/docker.sock
, which is not the case when presented with a non-elevated Rancher installation. Activating this profile will override theDOCKER_HOST
seen by these dependencies, pointing it instead atunix://$HOME/.rd/docker.sock
. - integration-test: Some integration tests require Docker and automatically start/stop Docker Compose services while executing tests (i.e. see the test/test-mda-models/test-data-delivery-pyspark-patterns module). Note that the Maven build does not build the Docker images directly. The images are built within the Kubernetes cluster to speed up development builds and save disk space.
The first step in creating a new project is to leverage Mavenās archetype functionality to incept a new Maven project that will contain all of your aiSSEMBLE component implementations - Data Delivery and Machine Learning pipelines as well as Path to Production modules.
Open a terminal to the location in which you want your project to live and execute the following command:
./mvnw archetype:generate \
-DarchetypeGroupId=com.boozallen.aissemble \
-DarchetypeArtifactId=foundation-archetype \
-DarchetypeVersion=<version number>
This command will trigger an interactive questionnaire giving you the opportunity to enter the following information (in order):
- groupId
- artifactId
- version
- package
- projectGitUrl
- projectName
-
For details on these fields refer to https://boozallen.github.io/aissemble/aissemble/current/archetype.html
-
For detailed instructions on adding a pipeline refer to (LINK COMING SOON) https://boozallen.github.io/aissemble/current-dev/add-pipelines-to-build.html
When executing the aissemble
build for the first time, you may encounter the following transient error when building
the test-data-delivery-pyspark-patterns
module:
:: problems summary
:::::: WARNINGS
[NOT FOUND ] org.apache.commons#commons-math3;3.2!commons-math3.jar (0ms)
==== local-m2-cache: tried file:/Users/ekonieczny/.m2/repository/org/apache/commons/commons-math3/3.2/commons-math3-3.2.jar
::::::::::::::::::::::::::::::::::::::::::::::
:: FAILED DOWNLOADS ::
:: ^ see resolution messages for details ^. ::
::::::::::::::::::::::::::::::::::::::::::::::
:: org.apache.commons#commons-math3;3.2!commons-math3.jar
::::::::::::::::::::::::::::::::::::::::::::::
:::: ERRORS
SERVER ERROR: Bad Gateway url=https://dl.bintray.com/spark-packages/maven/org/sonatype/oss/oss-parent/9/oss-parent-9.jar
SERVER ERROR: Bad Gateway url=https://dl.bintray.com/spark-packages/maven/org/antlr/antlr4-master/4.7/antlr4-master-4.7.jar
SERVER ERROR: Bad Gateway url=https://dl.bintray.com/spark-packages/maven/org/antlr/antlr-master/3.5.2/antlr-master-3.5.2.jar
If this occurs, remove your local Ivy cache (rm -rf ~/.ivy2
) and then manually download the dependency that failed to
download. Taking the above error message as an example, the following Maven command would download the needed commons-math3 jar:
./mvnw org.apache.maven.plugins:maven-dependency-plugin:get -Dartifact=org.apache.commons:commons-math3:3.2