AI-Augmented Software Modernization and Insight
Description
The IRS maintains a large portfolio of complex Java/Spring Boot applications that accumulate technical debt, dependency conflicts, and inconsistent coding practices. Manually identifying outdated libraries, researching upgrade paths, and resolving build failures consumes significant engineering time and slows modernization efforts. AI is intended to reduce this burden by automating research, detecting vulnerabilities, and proposing secure, standards-compliant code improvements.
Detailed example
Prioritized lists of technical debt items (e.g., outdated dependencies, deprecated APIs, security vulnerabilities). Suggested code refactorings and dependency upgrades. Generated draft documentation, test cases, and migration strategies. Analytical reports highlighting cross-repository trends in build failures and performance issues.
AI / analytics pattern
Generative AI: AI that generates new or synthetic content (e.g., images, videos, audio, text, code).
Automation level / stage
b) Pilot – The use case has been deployed in a limited test or pilot capacity.
Expected benefit
For the IRS mission: Faster modernization of taxpayer-facing systems, improved system reliability, and reduced risk from outdated or insecure software components. For the public: More resilient digital services, reduced downtime, and faster delivery of secure taxpayer tools and features. For IRS engineers: Less time spent on repetitive research, freeing focus for high-value mission logic and innovation.
Audit / financial statement impact
The output is not presumed to be high-impact and is not used as the principal basis for significant decisions/actions
Controls / human review
ATO: Yes; PIA: Not published
Data needed
Training: The model was trained on publicly available, licensed, or permissively shared text, code, and documentation, supplemented by human-authored examples for language, reasoning, and coding tasks. It was not trained, fine-tuned, or evaluated on IRS or other proprietary government systems, taxpayer data, or TaxPro code. All datasets were filtered to remove PII, secrets, and sensitive financial data in accordance with privacy and security policies. Fine-Tuning and Alignment: The model underwent supervised instruction tuning using de-identified and synthetic examples focused on software engineering, documentation, and compliance use cases. Reinforcement learning from human feedback improved quality, clarity, and policy adherence. Domain calibration relies on open-source Java/Spring projects, Maven build patterns, and general DevSecOps documentation, with no IRS-specific content. Evaluation and Governance: Performance was assessed using quantitative benchmarks for reasoning, code generation, and factual accuracy, including datasets such as HumanEval and MMLU, along with qualitative reviews for ethical, security, and accuracy standards. Periodic bias, privacy, and red-team testing support AI governance. All datasets undergo licensing and PII review. Operational prompts, logs, and outputs are not used to retrain the base model. Model documentation and provenance support auditability and align with federal AI risk-management frameworks, including NIST AI RMF 1.0.