Advanced Analytics Platform SOW.pdf

Advanced Analytics Platform (AAP) SOW

Outline/Table of contents:

1. Summary

2. Qualified Personnel

3. Project Management

4. Consultancy Services/Stakeholder Engagement:

5. Data Integration / Ontology Implementation Services:

6. Custom Application Development:

7. Analytics and Data Science

8. Cybersecurity Services:

9. Training and Education:

10. Ongoing Support and Maintenance:

Summary

FAA has recently acquired Palantir Foundry for use in it’s Enterprise Advanced Analytics Platform (AAP). This RFI is intended to understand services available to proficiently integrate and build solutions for various mission needs across the FAA. Responding vendors should be capable of providing implementation services to configure Foundry for the FAA’s mission needs. The contractor should demonstrate its ability to provide a project team of personnel who are sufficiently trained or certified on Palantir software products to successfully configure the solution, as well as have domain experience working the FAA’s safety data sources or equivalent. This project team will be responsible for the implementation, configuration, and management of the FAA’s instances of Foundry.

1. Qualified Personnel: The vendor should provide experienced personnel with the requisite skills and expertise to configure and manage Palantir software products in support of the FAA’s mission. The project team should demonstrate the following key skills and capabilities:

• Advanced data engineering, integration, and transformation techniques using the Platform’s engineering tool suite.

• Proficiency in data analysis and visualization tools to effectively interpret and present insights derived in the Platform.

• Understanding of machine learning models, algorithms, and statistical methods to leverage Foundry's advanced analytics capabilities and applying them to real-world problems.

• Familiarity with APIs and programming languages such as Python, R, SQL, Java, TypeScript/JavaScript, and C++ valuable for customizing and extending Foundry's functionality and integrating it with other systems.

• Strong engineering background and expertise in fields such as computer science, software engineering, and data science.

• Experience working with the FAA’s aviation safety datasets and source systems or equivalent experience with FAA data relevant to the use-case implementation of Palantir software.

2. Project Management: The contractor should provide a dedicated project team who will be responsible for configuring Foundry for the government’s needs, supporting ongoing operations of the Platform, and ensuring project outcomes are achieved. This project team is responsible for the implementation, configuration, and management of the FAA’s instances of the Platform and will provide the following project management and oversight activities:

• Provide comprehensive project management and oversight for the duration of the period of performance.

• Provide a project lead to serve as a primary point of contact between the government, the contractor team, and any subcontractors, as appropriate.

• Coordinate regular meeting cadences to include project check-ins, training sessions, and user feedback sessions.

• Document feedback from the government on configured Foundry capabilities and oversee the incorporation of user feedback into the solution.

• Oversee the implementation schedule, activities, and progress toward project outcomes.

• Proactively identify risks and mitigation strategies throughout the duration of the project, working with the appropriate government stakeholders to effectively resolve risks to attaining project outcomes.

• Ensure the timely delivery of all contract deliverables to the government

3. Consultant Services/Stakeholder Engagement: provides consultancy services to help organizations understand their data landscape, identify opportunities for improvement, and implement best practices in data management and analysis.

The project team should follow a close engagement model with the government that emphasizes attaining project outcomes, incorporating user needs and feedback, and prototyping in rapid and user-engaged cycles. Throughout the period of performance, the project team should actively engage government stakeholders and SMEs to ensure the configured solution is effective for users. The project team should support the following activities:

• Identify government users and SMEs who will provide subject matter expertise and engage in solution testing and feedback.

• Provide thought partnership to translate user needs into effective Foundry capabilities and work with the government to define success criteria for each use case.

• Lead and participate in regular touchpoints between the government and project team in order to ensure progress is being made toward desired outcomes.

• As capabilities are configured, hold testing and feedback sessions with identified government users and SMEs.

• Engage the government’s IT stakeholders to ensure compliance with IT standards and policies, provide architectural and security documentation, and support the government’s broader IT vision.

• Establish, as needed, a monthly Steering Committee with key government and contractor leadership to discuss the strategy of the solution, mitigate risk, and demonstrate progress and outcomes to date.

4. Data Integration Services/Ontology Implementation: assist FAA organizations in integrating data from disparate sources, ensuring that the data is clean, consistent, and ready for analysis. Make available its standard APIs, documentation, and all transformation logic for FAA data sources for use by FAA in order to facilitate data access, connectivity to other systems, and to identify derived data products created from FAA data sources. The project team should leverage the Foundry ontology to configure a unique object model for the FAA’s needs, including through the following tasks:

• Collaborate with government users and subject matter experts (SMEs) to understand the FAA’s most important concepts and business processes that will be reflected in the ontology and leveraged in user workflows.

• Design the optimal object model that supports the desired use cases and outcomes.

• Define the various object types as well as the actions that users can take on objects.

• Appropriately map and transform datasets to interpretable objects and establish the relevant linkages among objects to depict data relationships.

• Perform ontology review with key government stakeholders, SMEs, and business users to verify that the object model is implemented effectively and incorporate feedback into the ontology.

• Support the creation of new objects over time, as well as updates and enhancements to the existing objects and ontology.

• Configure front-end user interfaces to explore, filter, and interact with objects, as well as display important object properties and links to other objects.

• Configure the appropriate user permissions to update, change, and add to the ontology over time while maintaining an audit trail of changes to objects.

• Enable users to conduct queries and analysis on single or multiple objects of the same type, across an entire object type, and across multiple object types.

• Ensure use case dependencies to objects are mapped.

• Configure and monitor health checks and alerts on the ontology.

Data Versioning— vendors must store data and the metadata associated with a particular point in time, allowing users to recreate what the organization knew at any point in the past. This is important for historical purposes and for information or designation that are dynamic that change over time while other markers (e.g., aircraft serial numbers) remain constant or for request of information available to the FAA between certain dates or events.

Data lineage—Vendors must track dataset transformations across AAP applications and make available the dependencies and derivations from these transformations. By maintaining a record of the information and logic used to produce every output, the vendor must ensure reproducible results can always be tracked and data quality issues are fixed proactively.

Data traceability—Vendor must retain within AAP a detailed record of all data integration, including time of ingestion, source, and revision history.

5. Custom Application Development: For FAA business units with specific needs, vendors should demonstrate past performance in developing custom applications that are built on top of Palantir Foundry to provide tailored solutions. The project team should be capable of configuring front-end interfaces and workflows within Foundry in support of the government’s use cases. The project team should support the following workflow configuration activities:

• Define analytical and operational workflows, working in close collaboration with FAA stakeholders to understand business processes and user needs.

• Define the desired outcomes and success criteria for each use case. Success criteria may include but is not limited to successful user adoption rates, increased automation of manual tasks, enhanced data access and data usability, and accelerated time to insights.

• Configure Foundry applications and tooling to support the defined workflows and use cases, including data exploration, analytical, operational, reporting, and decision-making workflows.

• Provide guidance and expertise on Foundry capabilities to government users on the most effective implementation of workflows using either point-and-click tooling or code-based methods.

• Deliver configured interfaces to government users to test capabilities.

• Work in rapid prototyping cycles to refine initial workflows and incorporate government feedback.

• Configure integrations with additional third-party tools, as needed, to support dissemination of data and analytical products in Foundry to downstream systems.

6. Analytics and Data Science: provides tools and support for data analysis and the application of data science techniques, enabling organizations to gain insights and make informed decisions based on large datasets.

Reports and Dashboards

Vendors must generate and make possible the exports of reports, and live, highly configurable dashboards with pre-set calculations, filters, and drill-down capability;

allowing users to peruse both summary and detailed information as well as calculations of specific metrics that continuously update as new data enters the system.

Nontechnical and casual users, such as executives at the FAA, must be able to access the information with minimal training and view information through easy-to-digest dashboards. In addition to live, in-AAP reports, AAP solutions must allow users to easily share and export data. AAP solutions must use open API file formats and export data in a number of translatable ways, including XML, CSV, XLS, PPT, PDF formats, and other common file formats upon request.

Other Analytical Capabilities

Vendors must provide both technical and nontechnical users with a wide variety of highly configurable analytical workflows that provide proactive information and mission insights.

Nontechnical users should be able to perform broad analyses with front-end reporting via point-and-click operations enabling analytical users to write code in languages such as SQL, Python, R, Java, etc. and import any open source package.

Tabular— AAP solutions must allow for the use of collaborative spreadsheet applications and allow users to search and query datasets, apply standard cell-level functions, sync results for further analysis elsewhere in the platform, and save their results back to the platform.

Graphical—AAP solutions must allow technical and nontechnical users to conduct point-and-click analysis. It must allow users to examine a summary of massive-scale datasets and then isolate patterns and drill down into details of these datasets with a variety of platform generated visualizations, including histograms, statistical bar/line charts, heat maps, and more. Users must be able to save their work in the application and share their analysis, including any applied filters, directly within the platform. Users must be able to seamlessly duplicate their analyses to explore various hypotheses.

Proactive/Rule-Based—AAP solutions must feature automated alerting capabilities for user to improve decision-making and data quality, and give users faster access to information. Furthermore, AAP solutions must enable users to proactively identify invalid data through customizable business rules at a user-specific or organization-wide level. And must allow users to create alerts to individuals or groups via email, or other identified FAA notice in the platform, based on established conditional triggers.

Interoperability with Third-Party Tools

AAP solutions must be a modular architecture for ease of customization to current and future state workflows. Data integration with third party tools must include a robust collection of Representational State Transfer (REST) APIs to ensure that the platform is future-proof and interoperable with an organization's existing technology landscape. The platform must provide several “push”/”pull” mechanisms to external applications, including file transfer and common standard connections (e.g., common structured data set connectivity providers such as, Oracle, OleDB, ODATA, ADO.net, ODBC, JSON, and XML such as connecting to a SQL Server instance/Database). External systems must also “pull” from the platform via methods such as common drivers and REST Web Service APIs and by connecting directly to the Distributed File System where data lives.

ADVANCED ANALYTICS

The “Notebooks” application must give advanced analytical users an environment to perform technical analyses, build workflows, generate visualizations, and author or edit code in the programming language of their choice. Users must be able to integrate pluggable modules to let users write data transformation code in the language of their choice (Java, Python, R, SQL, etc.). Users must be able to also branch data and code within an isolated “sandbox” environment that ensures their work does not impact data across the platform until it is published to the production environment.

The “Notebooks” must support the testing and validation of existing and future state predictive models through the use of “model drift” adjustments. Having the capability to determine the stability of any given model, a tracking error feature must be enabled to quickly and effectively keep a model’s predictive capabilities within an acceptable ban of tolerance as defined by the user.

Users must be permitted to integrate in analytics off the central data and information source, without sourcing and manipulating data from disparate sources to begin their work. The platform must ensure the work of advanced analytical users creates an immediate impact on the workflows of less technical users and SME. “Notebook” users must be able to template their logic and translate it using defined input parameters to languages other than the one that they wrote their initial analysis, making their models accessible to other technical users and the rest of the organization. As outlined in the next section.

Machine Learning (ML)

Vendor must enable FAA advanced analytical users to build and test ML models on AAP, saving those models back to AAP. Enable analysis of massive quantities of data, while delivering faster, more accurate results in order to identify and leverage the broad range of open-source libraries (e.g., Spark ML, scikit-learn, TensorFlow, etc.) that must be available directly in AAP. Below are illustrative examples of advanced ML approaches the FAA may implement:

• Topic extraction algorithms—Models analyze high-volume unstructured data to highlight common issues and themes from numerous reports and statements such as engine inflight shutdowns trending and root cause.

• Supervised algorithms—Algorithms that support supervised learning through the use of historical data and labeled datasets to predict future events, training dataset, and a learning algorithm that produces an inferred function to make predictions about the dataset values. AAP ML solutions should provide targets for any new datasets after sufficient training. The algorithm can then be compared with the resulting dataset to find errors in order to improve the models accuracy.

• Unsupervised Learning Algorithms—Algorithms that support unsupervised learning that leverages datasets that are not classified or labeled and have the capabilities to find unnoticed structures in unlabeled datasets. The algorithms with these capabilities do not determine the “correct dataset,” but help drive inferences from datasets to descript the unnoticed structures in the unlabeled datasets.

• Combining Algorithms—that combine easily small-supervised label datasets with large unlabeled datasets. This semi supervised learning usually requires SME intervention to help train the model to be applied to the large unlabeled dataset.

• Environmental Interaction Algorithms—Algorithms that interacts with its environment by producing actions and discovers errors or rewards.

Trial and error search and delayed reward are the most relevant characteristics of ML. This method allows machines and software agents to automatically determine the ideal behavior within a specific context in order to maximize its performance. Simple reward feedback is required for the agent to learn which action is best; this is known as the reinforcement signal.

• Entity resolution systems—Models perform high-accuracy integration of incomplete and overlapping data sources to enable compliance with stringent reporting regulations

• Decision-support models— Models continuously learn from human expert judgment to speed up systematic defect identification and resolution.

“Notebooks” provide further differentiated value through:

• Iteration speed—“Notebooks” allow users to rapidly explore, view, and interact with data to refine transformation logic and derive useful outputs that are then extensible to other tools within Foundry.

• Reproducibility—“Notebooks” ensure that transformations are fully reproducible; rendering the transition from early-stage analytical work to a production-ready pipeline seamless.

• Collaboration—“Notebooks” enable disparate groups of users to share logic and outputs while working under comprehensive access controls and security protections.

Key features that enable the value of “Notebooks” include:

• An interactive console that allows for rapid exploration of data.

• A simplified interface to complex and domain-specific logic using transformation templates.

• Secure collaboration, lineage tracking, and branching workflows that make it possible for multiple users to collaborate on the same models and build on top of pre-existing work.

• An intuitive pipeline builder graphical user interface (GUI) that represents pipelines as graphs, nodes, and the relationships between data tables and transformations for a variety of user capabilities

(support staff, engineers, inspectors, SME, data scientists, managers, and leadership).

7. Cybersecurity Services: platforms are used for cybersecurity purposes, helping organizations to detect, prevent, and respond to threats.

Vendor must implement FAA security controls as defined on the FedRAMP website, applicable FAA Policy, and the Department of Transportation (DOT) Departmental Cybersecurity Compendium to operate securely based on the current DOT and FAA policy. This includes, but is not limited to, taking all appropriate protective measures to establish administrative, technical, and physical safeguards to ensure the confidentiality, integrity, and availability of information in all its forms and to comply with all laws, regulations, rules, and mandates. The contractor must also meet the following standards during performance of this contract:

• Contractor and Industrial Security Program, http://www.faa.gov/regulations_policies/orders_notices/index.cfm/go /document.i nformation/documentid/15573

• NIST 800.53 Security and Privacy Controls for Information Systems and Organizations, https://csrc.nist.gov/publications/detail/sp/800-53/rev-5/final

Single-Sign On (SSO)

The vendor must adhere to FAA authentication processes to allow a user to access multiple applications with one set of login credentials. It is required to enable two-factor authentication. This must follow security requirements provided through the FedRAMP security assessment for integration.

Access Controls

Vendors must establish and administer privileged user accounts in accordance with the individual’s role in relation to the system. All system access in AAP is role-based justified and limited using the least privilege principle. Privileged user accounts are verified annually and automatically purged following six weeks of non-use. Must support granular control restrictions of access to data depending on the specific data source.

The vendor solutions must allow FAA to apply highly granular permissions, including at the row level, thereby significantly decreasing inappropriate use of data. At the http://www.faa.gov/regulations_policies/orders_notices/index.cfm/go/document.i http://www.faa.gov/regulations_policies/orders_notices/index.cfm/go/document.i application level, AAP’s policy engine combines authentication and authorization functionality that must enable the FAA to create programmatically and enforceable access control policies, thereby ensuring that only authorized users are able to interact with the data allowed by the FAA's access restrictions. Permission levels must include discover, read, write, and owner. These access controls must also enable the transfer of responsibility between authorized managers and organizations. In addition, AAP solutions must enable the FAA to create and define custom hierarchies based on its organizational structure and restrict dataset access based on a user's hierarchical association.

The project team should support user enrollment and onboarding to include the following tasks:

• Support all necessary user enrollment and onboarding tasks for the approximate number of users agreed upon during the scoping of the statement of work.

• Integrate Foundry with the government’s active directory or SAML provider to inherit existing permissions and user group attributes.

• Implement all access controls and complete internal end-to-end testing to confirm their implementation.

• Enroll identified users and work with users to confirm access to the system.

8. Training and Education: offers training programs to help users understand how to use their platforms effectively and to develop the necessary skills to leverage big data analytics.

The FAA is seeking vendors to provide comprehensive AAP training and support solutions to all applicable FAA staff. AAP solutions must be intuitive and require minimal training for new users. Non-technical users must gain a working knowledge of the solution within just a few hours. Training will be designed to accommodate different formats, including remote or in-person user-centric classroom, online, on-call, face-to-face, and on-demand instructional video.

Training solutions must be tailorable to FAA's needs in several functional areas, including:

• Delivery—Vendors must provide training materials and be capable of both virtual and in-person training to FAA users, based on user role and profile. Virtual training must be conducted through online webinars. Three to five in-person group training sessions must be conducted with end users, on-site or virtually in a classroom format, or externally through Train-the-Trainer programs. FAA may also assign this training through FAA training platforms (eLMS or other).

• Duration and Frequency—In-depth training must be conducted at a duration and cadence necessary to identify, train, and onboard FAA SMEs, with additional training sessions being offered as required.

• Upon further configuration of the Platform, provide use-case specific training tailored to the unique configuration of FAA’s solution. This use case-specific training may include separate sessions broken out by user group roles and functions.

• Train users on successful operations of the configured Platform, including through options for training in a classroom setting, video teleconference, or side-by-side mentoring.

• Provide self-paced training and user enablement materials, such as general Foundry technical documentation and tailored workflow guides for configured use cases.

9. Ongoing Support and Maintenance: To ensure that their platforms continue to run smoothly, offers ongoing support and maintenance services to their clients.

SUPPORT SERVICES AND SYSTEM MAINTENANCE CAPABILITIES

Vendor must offer multiple support services options and system maintenance capabilities to ensure continuous access to technical expertise and resources. It must include:

Implementation Support

A vendor Project Team must implement, configure, monitor, and test all parts of the AAP solution for correct operation. A support team must provide support and work closely with core user groups, IT, etc., during initial installation and integration of the AAP solution. After initial deployment, the Contractor will coordinate support as appropriate and mutually agreed upon, including working with FAA to identify and train local knowledge experts who can provide effective supplemental support.

Online Support

The online support portal must give FAA users and stakeholders 24/7/365 access for support for security incidents, as well as access to relevant platform documentation (e.g., user guides, troubleshooting, etc).

Remote Support

In accordance with the Contractor’s standard support terms, the Contractor must provide direct access to the platform support Team by phone, email, video conference, or as necessary in person. General technical support must be available that is reachable by email or phone real-time to triage and troubleshoot issues. Technical support personnel work closely both with the customer and internal teams to maximize first-contact incident resolution.