New Local AI Simplifies Contracts

Overview

Existing user research reveals that customers often sign contracts without understanding their implications.

• An Adobe Acrobat Contract Survey from January, 2025 reveals that 70% of customers admit to signing contracts without knowing what’s in them. 32% of workers have faced negative consequences due to misunderstandings in contracts. The key factors behind this data point are that contracts are too long to read and the negative emotions surrounding contracts
• Another survey from the UK’s College of Law also reveals that 68% of participants admitted they either don’t read or don’t understand the contracts they sign. More than half of survey participants admitted they read their contracts to an extent but don’t always understand what they’re agreeing to. More than one in ten said they “hardly” or “never” read their contracts at all

Additionally, there are concerns with AI violating one’s privacy:

• A 2023 study by KPMG and the University of Queensland found three in four customers globally feel concerned about the potential risks of AI
• KPMG also found 63% of consumers were concerned about the potential for AI to compromise one’s privacy.
• A recent Pew Research Center also found that 81% of consumers think the information collected by AI will be used in ways people are uncomfortable with, as well as in ways that were not originally intended

In summary, the data reveals that AI can be a tool for analyzing contracts. However, there’s also the potential for AI to be a privacy concern as data may be sent to a company’s servers without a person’s knowledge. This suggests that there is a potential for the creation of a product that both:

• Allows for users to summarize and interact with their documents
• Also honors their privacy and promises to never send any confidential data to a company’s servers

My Role

I am the sole UX Designer who is using the following methods to propose a new UX for a customer-centric and open-source tool:

• Research & Discovery (Ongoing) ✅
• Analysis of Public Research ✅
• Analysis & Release of Survey Data ✅
• Workflow Diagrams ❌
• Lo-Fi Wireframes ❌
• Lo-Fi Prototype ❌
• Hi-Fi Mockups ❌
  
• Hi-Fi Prototype ❌

Timeline

• Foundational Deliverables: May 2025 – August 2025
  Lo-Fi Prototype: August 2025 – September 2025
  Hi-Fi Prototype: September - November 2025

Tools

• Penpot
• LibreOffice Docs
• Social Media Polls

01 - Target Audience

How stakeholders once viewed this proposal

• Users’ age between 20 and 60
  • Younger users aged 20 – 24 who are enrolled in an institution of higher ed
  • Middle-aged users from 30 – 40 are high-collar professionals in the industries of User Experience (UX), Graphic Design, Copywriting, Engineering, K-12 Education, and Government
  • Older users from 50 – 60 working in Office Management, Finance & Accounting, Recreational Services, Psychology, and Social Work
    
• Monthly budget is between $2000 – $8,000
• Internet Speeds
  • Mid-Tier: 500 Mbps Download, 10Mbps Upload
  • High-Tier: 1000 Mbps Download, 20 Mbps Upload

02 - Hardware Requirements

The app should be able to run on as many Mid to High range machines:

•  Mid-Range Machines
  • 4 to 8-core CPU
  • 6 GB Ram
  • 10GB available storage
  • Windows 10,11, macOS Ventura, Sonoma, Sequoia (currently supported versions), and Debian-based Linux Distributions
    

• High-Range Machines
  • 8 to 16-core CPU
  • 16 to 32 GB Ram
  • 25 GB available storage
  • Windows 10,11, macOS Ventura, Sonoma, Sequoia (currently supported versions), and Debian-based Linux Distributions

03 - Assumptions

• Privacy – To address the issue of privacy, a local Large Learning Model (LLM) that can be 2 to 4.5 GB in size will need to be downloaded to the user’s computer with full disclosure of the process and how a local LLM differs from a cloud-based LLM such as ChatGPT, Gemini, Docusign AI, Adobe, etc
• Learning Fast – AI is changing quickly — more AI models are expected to come in the near future, and the cons of each AI now will someday be addressed
• Consent – Users want to be confident that any document uploaded to an AI model will never be shared with anyone without consent
  Quality UX First over Politics – Users expect a friendly AI experience, regardless of whether it’s open source or proprietary. Although, for conscientious users, open source will be a selling point
• Competition with Proprietary Offerings – Microsoft’s Copilot AI is fully integrated with the Windows 11 OS, which will make it challenging to promote the benefits of local AI and privacy to this user base
• Competition with Open Source Offerings – RedHat's OpenShift AI is becoming a top feature in Red Hat Enterprise Linux, so targeting Debian-based distributions will be a key strategy for this project

04 - Constraints

• Voluntary Contributions – This will be an open source project, contributors will come and go as they please with no compensation
• Utilizing Alternative Channels – Only free methods can be used to attract new users and contributors. Other channels will need to be utilized. Potential options include Discord Chat, Facebook, X, and some Mastodon Instances that are free of charge

05 - Discovery

During my research of AI tools, I discovered GPT4ALL from Nomic AI, an MIT-licensed application for Windows, macOS, and Linux that’s designed to run local AI models on desktops and laptops. Before starting this deep dive of the first launch experience, I wanted to ask the community what the biggest concerns are for running local AI models:

• Received 7 responses to a poll, across both LinkedIn and Facebook, that asked What would be your biggest concern with running a local AI model on your PC/Mac? with the following options:Large file size/long download
  • Needing the best specs to run
  • High electric bills
  • Windows or macOS blocking it
  • Used data to inform the user flow, with pros and cons, for the first launch of GPT4ALL on Ubuntu Linux

The poll results reveal that 57.6% are concerned with large file sizes and long download times, 42.4% expressed concern with needing the best specs to run local AI models, and no participants expressed any concerns with high electric bills or the operating system blocking its installation. The results suggest that:

• Informing users of the size and estimated download time for each AI model is crucial
• Users need to be aware of the system requirements of each model to avoid running ones which could damage their machine

06 - User Persona (Deepseek-R1)

Alex

Age: 36

Gender: Male

Nationality: United States

Background: Alex is a 36-year-old IT professional who resides in the U.S. He has developed a strong reliance on AI for various aspects of his life, including personal productivity and decision-making. However, this reliance has been precipitated by his growing awareness of the ethical implications of AI.

Goals

• To maintain control over his private life and avoid sharing sensitive information online
• To use AI tools without compromising his autonomy or privacy

Concerns

• Loss of Control: Alex fears that his personal information may be misused or sold without his knowledge
• Fear of Compromise: He is vigilant about how his data might fall into the wrong hands, especially when using AI services

Compromises

• Limited Information Sharing: Alex prefers to share minimal personal details with online platforms and AI tools
• Trust in Medical Professionals: He feels comfortable sharing health information only with trusted medical professionals

07 - GPT4ALL Workflow Diagram

A workflow diagram was created for GPT4ALL at first launch to identify the current user experience and compare it against the existing research, poll results, and user persona.

Pros

• The explore models page offers great options, including Llama 3 8B Instruct and Deepseek-R1-Distill-Qwen-7B, to choose from
• Important data points including file size, RAM required, Parameters, Quant, and Type are included for each model
• Warnings are given to models not recommended for the user’s specs
• Users can also link GPT4ALL to cloud providers like OpenAI and Mistral
• HuggingFace models can be searched for and downloaded as well

Cons

• The Welcome modal asks users if they would like to opt-in for anonymous usage-analytics and/or sharing of chats to the GPT4ALL datalake
• The release notes conflict with the more important information about the usage analytics and the datalake
• Unlike GPT4ALL, the datalake is under the Apache-2.0 license, and its purpose is to “ingest, organize and efficiently store all data contributions made to gpt4all”
• Latest News takes up half of the horizontal space of the main dashboard, which the poll results suggests may not be as important for users
• If GPT4ALL is being launched for the first time, then the default dashboard may be overwhelming for some users

08 - Next Steps

• Create separate workflow diagrams for the AI Chat, LocalDocs, Installed Models, Explore AI Models, and Settings pages
• Create new user flow diagrams that improve the experience with interaction notes and changes
• Create lo-fi wireframes for prototype

Copyright © Nathan Nasby