AI-Augmented Software Modernization and Insight

Overview

Development stage

PilotSource: b) Pilot – The use case has been deployed in a limited test or pilot capacity.

High-impact designation

Not high-impactSource: c) Not high-impact

Impact justification

The output is not presumed to be high-impact and is not used as the principal basis for significant decisions/actions

Start date

2025-08-20 (day precision)Source: 8/20/2025

Withheld from public reporting

NoSource: a) No

Mission

Topic area

Information Technology

Operational functions

No function tag matched Derived by keyword rules; see methodology

Problem the AI is intended to solve

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 Manually identifying outdated libraries, researching upgrade paths, and resolving build failures consumes significant engineering time and slows modernization efforts.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.

Expected benefits

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the IRS mission: Faster modernization of taxpayer-facing systems, improved system reliability, and reduced risk from outdated or insecure software components.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.

System outputs

Prioritized lists of technical debt items (e.g., outdated dependencies, deprecated APIs, security vulnerabilities).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.

Technology

AI classification

Generative AISource: Generative AI: AI that generates new or synthetic content (e.g., images, videos, audio, text, code).

System name(s)

Not reported

Custom-developed code

Yes

Public source code

Not reported

Sourcing

How it was built

Developed in-houseSource: b) Developed in-house

Vendor (as reported)

Not reported

Vendors (standardized)

None on the standard list

A vendor is the supplier named by the agency. It does not identify the underlying model or AI technology.

Data and privacy

Involves PII

No

Privacy Impact Assessment

Not reported

Authorization to Operate

Yes

Demographic features

k) None of the above

Training and evaluation data

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.

Federal Data Catalog

Not reported

Governance

0 of 8 minimum-practice questions answered. A blank answer means the agency reported nothing; it does not mean the practice is absent.

Potential impacts and how they were identified

Not reported