Paperless Parts has created purpose-built AI for manufacturing, including models trained specifically to understand technical drawings, RFQ packages, and the nuances of how custom parts are specified, quoted, and produced. Unlike general-purpose AI tools, our models and workflows are designed from the ground up for the way manufacturers actually work. We deliver these capabilities while maintaining the most stringent security and compliance requirements in the industry (including CMMC and FedRAMP) and protecting our customers’ proprietary information.

Paperless Parts values our customers’ trust above all else. As such, we have adopted the following set of guiding principles:

Our AI Principles

Secure & Compliant

As a CMMC-compliant Cloud Service Provider with a FedRAMP Moderate Equivalency, protecting your data is our highest priority. Guided by FedRAMP and NIST standards, we maintain strict controls across our organization, computing infrastructure, and applications, including the AI features within our platform and the R&D processes used to develop our manufacturing-aware AI models. Our AI-powered features are safe for use by customers with CMMC Level 2 requirements.

Most AI approaches require a choice between capability and data protection. We engineered a different path. Our patent-pending, privacy-preserving AI architecture is designed specifically for environments where data sensitivity is non-negotiable — including Controlled Unclassified Information (CUI). This approach has been independently peer-reviewed and assessed: our paper, “Privacy-Preserving AI for Document Understanding with Controlled Unclassified Information,” was presented at the 2024 IEEE High Performance Extreme Computing Conference, and the implementation of our AI features is continually reviewed as part of our annual, third-party FedRAMP assessments.

Note: AI features in our products, whether powered by Paperless Parts technology or third-party providers, are subject to the same compliance requirements as the broader product. Aerospace & Defense customers should refer to the Shared Responsibility Matrix for more information.

Your Data Stays Yours

Your customer, pricing, and shop data are private to you. Your pricing formulas, margins, material costs, labor assumptions, internal notes, and information about your customers are never shared with or exposed to other customers.

Our platform is architected so that your data is isolated from other customers’ data. No AI model produces a suggestion for another customer that exposes your data — not your pricing, your quoting history, nor your technical configurations. If multiple customers upload the same or a similar part, the system does not share, infer, or reveal pricing between them. Each customer’s quoting experience is driven by their unique account configuration and private data.

Transparent & Human-Centered

Our AI features are designed to assist, not replace, human judgment. AI inside Paperless Parts does not generate or suggest shop-specific pricing or optimization strategies. We believe each shop approaches pricing and costing decisions differently based on its specific capabilities and business strategies.

The platform clearly identifies when AI-generated suggestions are being made and gives users full control to accept, modify, or reject them. Final decisions always remain with the user.

We are committed to making AI suggestions not only visible but understandable. When Wingman identifies a specification, dimension, or requirement, it indicates the source (the document and location where that information was found) so estimators can verify suggestions against the original print and apply their expertise with confidence.

Fair & Consistent

We are committed to ensuring our AI features perform consistently and reliably across the full range of inputs our customers provide. We actively test and monitor accuracy across different print qualities, drawing conventions, and document structures to identify and address performance gaps.

When our models perform better on certain input types than others, we treat that as a problem to solve, not a limitation to accept. AI will never be perfect, and some technical data packages are easier to read than others. However, estimators across our customer base deserve the same quality of AI assistance, regardless of how their technical documents are formatted or generated, and we strive to perform well on a wide variety of data.

Additionally, as state-of-the-art AI systems get more and more capable, they are also becoming increasingly expensive. We strive to keep AI-supported assistance accessible and affordable to manufacturers.

Accountable & Governed

Responsibility for the development, performance, and safety of our AI capabilities rests with our engineering and product leadership teams. We maintain documented governance processes for how AI models are developed, tested, deployed, and monitored.

If you believe an AI feature has produced a problematic suggestion, or if you have questions about how AI operates within Paperless Parts, you can reach our team at [email protected]. We are committed to reviewing reported issues promptly and transparently.

We also monitor the evolving AI regulatory landscape — including the EU AI Act, emerging U.S. state-level AI legislation, and updates to NIST AI Risk Management Framework guidance — and will adapt our practices as requirements develop that are relevant to our customers and their supply chains.

Continuously Tested & Improved

We maintain a rigorous process for developing and deploying AI capabilities. All features are tested against defined benchmarks prior to release (validated on real-world manufacturing documents, not synthetic test data) and we continuously evaluate performance, gather customer feedback, and refine our models to ensure they deliver meaningful and reliable value in production quoting workflows.

Our AI models are versioned and tracked. When we update or refine a model, we validate the new version against our accuracy benchmarks before deployment. We monitor for performance degradation over time, and we will retire or replace models that no longer meet our quality standards rather than allow accuracy to silently decline.

How We Use Data to Improve Our Platform

Paperless Parts uses a limited set of customer data, in aggregated and abstracted form, to improve the overall performance of platform AI features. This section explains what we use, what we never use, and how we maintain control.

What we use: The foundation of our custom AI models is synthetic datasets created based on our team’s expertise. We have used a random sample of technical drawings uploaded to Paperless Parts to train proprietary models that automatically detect and interpret the symbols, patterns, and structures specific to manufacturing documentation, including GD&T symbols, BOM table layouts, specification blocks, and file naming conventions. This is the foundation of what makes our AI manufacturing-aware rather than general-purpose: models trained on real-world manufacturing data, applied in controlled, non-identifiable ways. 

What we never use: We do not train any models to be used in generative AI applications using customer data, thereby eliminating the risk of confidential data exposure through AI features.

Third-party providers: Where AI features incorporate third-party services as part of our processing pipeline, those providers are subject to the same compliance requirements as our broader platform. Customer data processed by third-party AI providers is not stored by those providers beyond what is required for the immediate processing request, and is never used to train their models. A current list of AI-related sub-processors is available upon request.

Opt-out: If you prefer that your data not be included in any aggregated model improvement processes, contact us and we will honor that request. Your use of Wingman and other AI features will not be affected.

FAQ

What is Paperless Parts Wingman™?

Paperless Parts Wingman is a suite of manufacturing-aware AI capabilities embedded directly in the quoting workflow. Wingman combines advanced document understanding — including automated extraction of GD&T callouts, tolerance specifications, material requirements, BOM structures, and RFQ terms — with contextual autofill that surfaces the objective, technical information estimators need to set up a quote accurately.

Rather than requiring estimators to manually parse multi-page print packages for easy-to-miss details, Wingman reads and interprets technical documents the way an experienced estimator would, then presents its findings for human review and decision. Estimators can then get to what they do best faster: leveraging their expertise. The presence of a purple graphic denotes Wingman offering a suggestion or recommending an action.

Refer to our Knowledge Base article to learn more or view up-to-date, specific examples of AI-supported automation currently available in the platform.

Who has access to my data?

The security of your information is paramount, and we develop and deploy all of our models within our FedRAMP boundary and in a CMMC, NIST, and FedRAMP Moderate-compliant manner. Paperless Parts has strict data controls in place that ensure your data is not shared with other customers or anyone other than authorized Paperless Parts team members, all of whom are US Persons.

How accurate are your AI models?

We measure and validate the accuracy of every AI model within our platform against defined benchmarks, and we use customer feedback on AI-powered suggestions to continuously improve. We do not guarantee the accuracy of suggestions due to the inherent limits of AI, and we stress the importance of human review of all AI suggestions.

Wingman’s document understanding pipeline goes beyond standard text extraction. It combines optical character recognition with manufacturing-specific models trained to identify and interpret the structures that matter in technical drawings — from GD&T symbols and feature callouts to BOM tables and specification blocks. This layered approach means the system doesn’t just read text; it understands the context and structure of a print the way an estimator would.

Accuracy depends on input quality — the system performs best on cleanly exported PDFs and may not work well on hand-written or heavily degraded documents. We are actively working to expand the range of input formats and quality levels our models handle effectively.

For customers interested in understanding the accuracy profile of specific AI features, we are happy to share more detailed information. Contact your account team or reach out to [email protected].

How do you protect sensitive data (including CUI) from unauthorized disclosure when used with AI features?

Generative AI models are a good solution to certain problems in manufacturing, but most approaches require sensitive data to flow through general-purpose models, creating unacceptable risk for defense and aerospace environments. Paperless Parts took a fundamentally different architectural approach: our privacy-preserving AI extracts information from technical documents in ways specifically designed to prevent data leakage, without sacrificing capability. We do not provide generative output to one user that has been trained on another user’s data.

This approach has been peer-reviewed and presented at the 2024 IEEE High Performance Extreme Computing Conference. You can read our paper, “Privacy-Preserving AI for Document Understanding with Controlled Unclassified Information,” for technical details.

Updated: May 2026

The AI landscape is evolving fast, and so are we. What won’t change is our promise to you: to use these technologies responsibly, transparently, and always in the interest of helping your business compete and win.

Questions? We’d love to hear from you at [email protected].

For additional questions on the use of AI within Paperless Parts, please contact [email protected].

Originally Published: February 12, 2024
Updated: May 1, 2026