Treasury Readability

Overview

Development stage

In development or acquisitionSource: a) Pre-deployment – The use case is in a development or acquisition status.

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

Not reported

Withheld from public reporting

NoSource: a) No

Mission

Topic area

Procurement and Financial Management

Operational functions

Document processing and extraction, Procurement and acquisition Derived by keyword rules; see methodology

Problem the AI is intended to solve

Government contract descriptions are indecipherable to the average American. Treasury Acquisition Procedures Update No. 25-05 requires that contract descriptions in the Federal Procurement Data System provide a clear and concise plain language description of products or services acquired under a specific contract. Under the Federal Funding Accountability and Transparency Act (FFATA) et seq, the IRS must publish contract spending reports for viewing by policymakers and the public. The IRS has previously funded and published research on the use of models to validate the accuracy of structured contract spending data such as dates, dollar amounts, and addresses. Advances in large language models have made it feasible to automatically assess the quality of unstructured text descriptions of contract purchases. This coding provides automated suggestions from a Large Language Model that outlines failings and details suggestions on how contract descriptions can be improved. AI prompts are used to evaluate the quality of contract descriptions on an adjectival rating scale and provide narrative feedback on how to improve descriptions.

Expected benefits

Help Treasury/IRS improve the quality of contract description language and increase transparency with agency contract spending. Evaluate the quality of contract spending descriptions on an adjectival rating scale. Generate narrative feedback on how to improve specific contract descriptions using AI. Generate reporting quantifying trends in contract description quality (e.g. improvement or decline in data quality).

System outputs

The developer will design prompts and deliver responses from an open-weight, open-source Large Language Model (LLM) to the client using an R-based workflow. Prompt refinement and model outputs will be iteratively reviewed and adjusted until the client confirms satisfaction with the responses. Example Model Outputs: Rating: Acceptable. The description “THOMSON REUTERS ANNUAL TAXATION SUBSCRIPTION” clearly identifies a subscription-based contract with Thomson Reuters. While concise, it lacks details about the specific services included, intended users, and usage context, which would improve clarity and understanding of the contract scope. Rating: Marginal. The description “BPA SET-UP FOR ISS ENTERPRISE SERVICE MANAGEMENT SOLUTION SERVICES” is unclear and relies on unexplained acronyms, which may confuse readers. It also omits key context such as the purpose of the service, intended users, and usage location, making the contract scope difficult to interpret and resulting in a marginal readability assessment.

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

Not reported

Public source code

Not reported

Sourcing

How it was built

Not reported

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

Not reported

Privacy Impact Assessment

Not reported

Authorization to Operate

Not reported

Demographic features

Not reported

Training and evaluation data

Not reported

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