Enterprise AI / Product Design / Frontend Engineering
Designing an AI-Assisted Legal Document Workspace
Exploring how structured inputs and conversational editing could support the generation and revision of complex legal documents.
Project statusWorking prototype — not launched to production.
Overview
A complex document. A new interaction model.
A large U.S. law firm explored using generative AI to help produce Private Placement Memoranda (PPMs), lengthy investment documents containing recurring structures alongside deal-specific information.
The team developed a prototype combining a guided intake questionnaire, AI-generated document drafts, and an interactive workspace for reviewing and revising generated content.
Interactive reconstruction
From a draft to a considered revision.
Interactive reconstruction based on a previous client prototype. Fictional content is used for demonstration.
The original project explored multiple interaction models. An earlier iteration allowed conversational instructions to update document content directly. A later iteration presented proposed text for user approval before applying it. This demonstration uses that review-and-approve interaction.
- Provide an instruction
- Generate a proposed revision
- Compare with the original
- Accept or reject the change
Use the sample instruction supplied in the assistant for this demonstration. The document changes only when you accept a proposal.
Private Placement Memorandum
3. Risk factors
An investment in the Company involves substantial risk. Prospective investors should consider the following discussion together with the other information in this fictional memorandum.
Membership interests are illiquid. Investors should be prepared to hold their interests for an extended period. The Company expects to pursue a sale of its properties when market conditions are favorable.
Fictional example for interaction demonstration. Not an offering or an original client document.
Select a section, then generate a sample revision to review.
My role
Connecting legal requirements and the frontend experience.
- Collaborated directly with a senior legal partner to understand document requirements and recurring patterns.
- Helped translate those requirements into a structured questionnaire.
- Designed and implemented the frontend experience.
- Worked with engineers responsible for AI generation and backend integration.
- Participated in iterative reviews and stakeholder demonstrations.
The challenge
Recurring structure, transaction-specific detail.
PPMs contain substantial structured content, but each transaction requires specific information and careful review.
The team explored whether AI could generate an initial draft from structured inputs while allowing legal professionals to inspect and revise the result.
Experience 01
Structured intake
The document-generation process required extensive structured inputs. The team organized the questionnaire as a multi-step wizard, using progressive disclosure and Back/Next navigation to collect the transaction-specific information needed for an initial PPM draft.
A senior legal stakeholder found the number of questions burdensome, creating tension between completeness and usability. The team explored AI-assisted prepopulation using information already supplied to help fill additional fields where possible and reduce repetitive entry.
Suggested values remained subject to user review and editing. The design sought to reuse known information while preserving user verification and enough detail for document generation.
Illustrative step 2 of 4
- Project setup
- Company information
- Offering details
- Review
Company information
Provide the company details that will inform the initial document draft.
- Legal company name AI-assisted suggestion
- Aster Vale Ventures, LLC
- Suggested from previously supplied information. Review and edit before continuing.
- Entity type
- Limited liability company
- State of formation
- Colorado
- Primary contact
- Morgan Ellis
Suggested values remained reviewable and editable in the original workflow.
Experience 02
Document workspace
After generation, the application displayed a multi-page document in a central reading and editing area. The workspace combined familiar document-editing conventions with conversational AI interaction.
Persistent document navigation helped organize lengthy generated content and kept sections accessible during review. A central document surface kept the artifact itself visible for reading and editing, with conversational controls alongside it rather than in a separate destination.
The project explored both direct conversational updates and a later review-and-approve model. Presenting proposed revisions before applying them gave legal professionals an explicit point to review and authorize a change to the document. The reconstruction above demonstrates that interaction; it does not reproduce every editing feature of the prototype.
Left navigation
- Document outline
- Page navigation
- Thumbnail-style page previews
Center workspace
- Generated document pages
- Scrollable document view
- Direct editing interactions
- Contextual controls for tables, layouts, and other document elements
Right-side controls
- Conversational AI assistant
- Requests to revise document content
- Changes applied directly to the generated document rather than merely returning text for manual copying
Experience 03
Stakeholder iteration
The prototype underwent multiple iterations involving engineers and legal stakeholders. The team demonstrated the experience to senior legal stakeholders.
Challenges & outcome
Generation quality and responsiveness remained barriers to production.
The prototype demonstrated the intended interaction model, but document-generation quality and responsiveness remained significant challenges. The initiative was ultimately discontinued without a production launch.
Reflection
Designing for the reliability of the whole system.
- Structured inputs can help constrain complex generative tasks.
- AI-generated documents still require human review and correction.
- Conversational editing benefits from direct integration with the artifact being edited.
- A compelling interface does not compensate for insufficient generation reliability.
- Production readiness depends on the quality of the entire system, not only the frontend.
Supporting engagement
A separate, deployed internal RAG assistant
In another project within the same client engagement, the team designed and developed an internal RAG-based assistant that could retrieve information from approved internal resources.
This assistant was launched and expanded beyond its initial limited rollout. It was a separate project from the discontinued PPM prototype.
My contributions
- Frontend implementation and UI iterations
- Collaboration with AI/backend engineers
- Prompt-related refinement
- User testing and feedback
- Participation in security-focused testing
- Creation of user guidance and prompting documentation