Artificial Intelligence-enabled Diligence Evaluation - AIDE

Problem Statement

Currently, business intelligence assessments are labor-intensive and prone to inaccuracies from manual efforts in capturing and analyzing key data points. The use of generative artificial intelligence (AI), particularly large language models (LLMs), offers potential for automating report generation, summarizing complex data, and creating consolidated dashboards for better decision-making. However, integrating this technology into existing program management workflows presents significant hurdles, including ensuring accurate data integration, connecting disparate proprietary and open-source data sources, and maintaining effective user engagement.

The Department of Defense (DoD) identifies and analyzes risks from individuals, entities, technologies, and transactions when assessing potential commercial vendors. Due to large volumes of data from multiple sources, timely and accurate generation of risk assessments and sharing of information across organizations and with senior leadership poses a unique challenge. 

Desired Solution Attributes

The Defense Innovation Unit (DIU) seeks solutions that leverage AI to streamline industrial foreign influence analysis, business intelligence, and due diligence by developing a platform and/or service that can efficiently process financial, business intelligence, and workflow data from multiple sources, providing detailed analysis, risk assessments, and summary reports which enhance the speed, quality, and effectiveness of due diligence processes. The solution should scan and assess a broad range of commercial data, applying a risk-based analysis framework to identify various potential vulnerabilities, including adversarial capital, intellectual property protection, supply chain resilience, business relationships, technology-sharing agreements, financial stability, cyber hygiene, and more. The platform should present insights on a professional dashboard for easy sharing and reporting to senior leadership.

Desired solution attributes include:

• Understanding of the problem sets and expertise in intelligence, business, and financial data sources and experience with delivery of business intelligence assessments or services.
• Data aggregation and integration to support proprietary, open-source, structured, and unstructured data, with scalable extract, transform, and load (ETL) processes, API integrations, and real-time data processing.
• Data processing and transformation including normalization, cleaning, and entity recognition to convert unstructured data into actionable insights, including support for various formats (JSON, CSV, XML, PDF, etc.).
• Generative AI integration with the use of large language models (LLMs) for text generation, summarization, and contextual understanding, with the ability to fine-tune models for specific datasets and industry knowledge.
• Analytics and reporting that implement advanced analytics for trend analysis, customizable scoring algorithms, automated report generation, and visualization tools.
• UI/UX: Customizable dashboards, real-time data refresh, and interactive charts for dynamic data exploration, with accessibility and collaborative features for human-AI interaction; design for user-friendly workflows, dashboards, and reporting.
• API integration and interoperability with full support for external systems, including RESTful and GraphQL APIs, secure authentication, and event-driven architecture.
• Security and compliance featuring end-to-end encryption, role-based access control, and compliance with applicable regulations.
• Scalability and performance represented in cloud-native, high-availability architecture with load balancing, low-latency processing, and fault tolerance.
• Data governance and quality assurance that features data lineage tracking, metadata management, automated anomaly detection, and data versioning.
• Flexibility for future integration of modular architecture that allows for future AI model integration and compatibility with open-source tools.

A fully functional prototype must:

• Integrate agreed-upon data sets with data provided by the vendor using proprietary, purchased, shared, transferred, accessed, licensed, sub-licensed, or other third-party data sources.
• Apply weighted risk metrics to implement a risk analysis or company viability scoring system.
• Support point-in-time risk assessments for report generation.
• Support search queries and continuous monitoring for collaboration in the due diligence process. 
• Display findings on a dashboard and allow extraction of results into presentations.

Submission Requirements 

Submissions should include an overview and technical details of the proposed solution. Inclusion of examples of the successful deployment of similar solutions in the commercial sector is highly encouraged.

Preference will be given to submissions that present a comprehensive and compelling solution to the problem statement and product requirements. Proposals should identify whether the submitter will employ partners or subcontractors and, if so, which companies would deliver which capabilities.

The Government requires cross functional collaboration and integration of technical outputs to ensure mission success. The Government intends to utilize this AOI in singularity, as well as a component of a more complex program, for which other AOI’s may be issued. Therefore, all solutions submitted in response to this AOI may be used to support, including technology insertion, into other prototyping efforts advertised by DIU. All AOIs will be governed by Commercial Solutions Opening (CSO) HQ0845-20-S-C001 dated 23 March 2020. 

Responsible AI (RAI)

All projects will be expected to comply with DIU's Responsible AI Guidelines and other DoD AI frameworks. Companies are encouraged to read the guidelines and develop a plan for implementation, which may be presented as part of the submission. Particularly relevant issues for this project include:

• Transparency and explainability of AI-generated content and recommendations
• Mitigation of potential biases in training data or model outputs
• Approaches for human oversight and ability to override AI systems
• Security and privacy protections for sensitive planning information
• Ethical considerations around AI's role in consequential military decision-making

The Statement of Work will require that the guidelines are followed in order to be eligible for a letter of completion and transition.