System Requirements Document

Artificial Inventor: AI-Powered Patent Filing Chatbot

Table of Contents

1. Introduction
   1.1 Purpose
   1.2 Scope
   1.3 Intended Audience
   1.4 Definitions and Acronyms
2. Overall Description
   2.1 Product Perspective
   2.2 User Classes and Characteristics
   2.3 Operating Environment
   2.4 Assumptions and Dependencies
3. Functional Requirements
   3.1 User Account and Subscription Management
   3.2 Conversational AI Mentor
   3.3 Patent Topic Research and Prior-Art Search
   3.4 Patent Document Drafting
   3.5 Claim Examination and Patentability Analysis
   3.6 Team Collaboration and Discussion
   3.7 Automated USPTO Filing
   3.8 Trademark Assistance and Multi-Jurisdiction Support
   3.9 Pro Tier: Multi-Model Verification
4. Non-Functional Requirements
   4.1 Performance
   4.2 Scalability
   4.3 Security and Privacy
   4.4 Reliability and Availability
   4.5 Usability
   4.6 Compliance
5. AI Model and Infrastructure Requirements
   5.1 Model Selection
   5.2 Fine-Tuning Approach
   5.3 Inference Infrastructure
   5.4 Cloud Backup and Redundancy
6. External Interface Requirements
   6.1 Bing Search API Integration
   6.2 USPTO Integration
   6.3 Third-Party LLM APIs (Pro Tier)
   6.4 Payment Gateway
7. User Interface Requirements
8. Subscription Tier Feature Matrix
9. System Architecture Overview
   9.1 Frontend Layer
   9.2 Backend Layer
   9.3 AI / ML Layer
   9.4 Data Layer
10. Constraints and Limitations
11. Future Enhancements
12. Appendix
    12.1 Requirement Traceability Matrix
    12.2 Revision History

1. Introduction

1.1 Purpose

This Software Requirements Specification (SRS) document defines the functional and non-functional requirements for the Artificial Inventor chatbot — an AI-powered conversational patent assistant. The system is designed to democratise the invention and patent filing process by mentoring users, particularly youth and independent inventors, through ideation, patent document drafting, prior-art research, and automated filing with patent offices.

1.2 Scope

The Artificial Inventor chatbot is a vertical SaaS application that will operate as the primary software product of the Cloud Wash Laundry platform. The chatbot will be deployed on locally hosted GPU infrastructure and will provide end-to-end patent assistance including topic research, WIPO-compliant document generation, team collaboration, claim examination, and automated USPTO filing. The system will offer two subscription tiers: a Basic tier aimed at youth and casual inventors, and a Pro tier for serious inventors and professionals.

1.3 Intended Audience

This document is intended for the project stakeholders, the ALGO development team, the Head of AI, the Cloud Wash leadership, and any third-party collaborators involved in AI model fine-tuning, infrastructure setup, or patent domain expertise.

1.4 Definitions and Acronyms

• USPTO: United States Patent and Trademark Office
• CIPO: Canadian Intellectual Property Office
• WIPO: World Intellectual Property Organization
• LLM: Large Language Model
• LoRA: Low-Rank Adaptation (fine-tuning technique)
• GPU: Graphics Processing Unit
• SaaS: Software as a Service
• MVP: Minimum Viable Product
• CDN: Content Delivery Network
• API: Application Programming Interface
• RAG: Retrieval-Augmented Generation
• RBAC: Role-Based Access Control

2. Overall Description

2.1 Product Perspective

The Artificial Inventor chatbot is the flagship software product within the Cloud Wash Laundry ecosystem. While the broader platform combines GPU compute infrastructure with car wash and laundry operations for heat reuse, this document focuses exclusively on the chatbot application. The chatbot will run on a locally deployed open-source LLM, served from on-premise GPU infrastructure, with a cloud backup for redundancy. It interfaces with external services including the Bing Search API for topic research, the USPTO website and database for prior-art searching and automated filing, and third-party LLM APIs for the Pro tier’s multi-model debate capability.

2.2 User Classes and Characteristics

2.2.1 Youth Inventors (Primary)

• Target Age: 10–17 years old
• Needs: Engaging, simple, and educational interface
• Features: Parental or guardian consent and oversight

2.2.2 Independent Inventors (Basic Tier)

• Target Audience: Adult hobbyist inventors and creative individuals
• Needs: Affordable tools for exploring patentability, drafting documents, and filing patents

2.2.3 Professional Inventors (Pro Tier)

• Target Audience: Serious inventors, small business owners, and professionals
• Needs: High-accuracy patent documents, multi-model verification, advanced collaboration features

2.2.4 Team Collaborators

Users invited to a team workspace to collectively brainstorm, discuss, and co-draft patent documents. These may include co-inventors, mentors, educators, or advisors.

2.3 Operating Environment

• Web-based Application: Accessible via modern browsers (Chrome, Firefox, Safari, Edge)
• Mobile-Responsive Design: Optimised for tablet and smartphone usage
• Backend Infrastructure:
  • Primary: On-premise GPU cluster (4x NVIDIA B300 GPUs)
  • Secondary: Cloud backup for redundancy
• External Integrations:
  • Bing Search API
  • USPTO electronic filing system
  • Third-party LLM APIs

2.4 Assumptions and Dependencies

• USPTO electronic filing system (EFS-Web / Patent Center) API or web automation interface remains accessible and stable.
• Bing Search API provides reliable access for topic and prior-art research.
• On-premise GPU infrastructure (4x NVIDIA B300 GPUs) will be operational before production deployment.
• An open-source LLM (from the recommended set) is available for fine-tuning with LoRA adapters.
• Mac Mini cluster is available for cost-effective inference serving.
• Users of the automated filing feature must provide valid USPTO credentials and accept responsibility for submissions.

3. Functional Requirements

3.1 User Account and Subscription Management

• FR-1.1: The system shall support user registration via email, social login (Google, Apple), and parental-consent-gated registration for users under 18.
• FR-1.2: The system shall support two subscription tiers: Basic and Pro, each unlocking different feature sets as defined in Section 8.
• FR-1.3: The system shall integrate with a payment gateway to handle monthly recurring subscriptions, upgrades, downgrades, and cancellations.
• FR-1.4: The system shall provide a user profile and dashboard showing active patents in progress, past filings, team memberships, and subscription status.
• FR-1.5: The system shall implement parental controls for youth accounts, including activity visibility, content filtering, and spending approval.

3.2 Conversational AI Mentor

• FR-2.1: The chatbot shall engage users in a guided conversational flow to help them articulate their invention ideas, identify unique aspects, and understand the patent process.
• FR-2.2: The chatbot shall adapt its tone, vocabulary, and complexity based on the user’s age group and expertise level (youth, beginner, professional).
• FR-2.3: The chatbot shall proactively ask clarifying and expansive questions to elicit the level of detail required for patent documentation, avoiding summarisation and instead expanding on user inputs.
• FR-2.4: The chatbot shall provide educational explanations of patent concepts (claims, prior art, obviousness, novelty, disclosure requirements) in age-appropriate language.
• FR-2.5: The chatbot shall maintain persistent conversation history per project, allowing users to resume sessions across devices.
• FR-2.6: The chatbot shall support brainstorming sessions where users can explore new invention ideas and emerging technologies interactively.

3.3 Patent Topic Research and Prior-Art Search

• FR-3.1: The system shall integrate with the Bing Search API to perform real-time web research on topics related to the user’s invention, retrieving relevant scientific papers, articles, product information, and technology overviews.
• FR-3.2: The system shall integrate with the USPTO patent database (PatFT and AppFT) to search for existing patents and published applications related to the user’s invention.
• FR-3.3: The chatbot shall present research results in a structured, digestible format, highlighting potential overlaps with existing patents and suggesting differentiation strategies.
• FR-3.4: The system shall support keyword-based, classification-based (CPC/IPC), and semantic search across patent databases.
• FR-3.5: The chatbot shall use retrieved research and prior-art data as context (via RAG) to guide the user through drafting and to improve the specificity and compliance of generated documents.
• FR-3.6: The system shall allow users to save, annotate, and organise research results within their project workspace.

3.4 Patent Document Drafting

• FR-4.1: The chatbot shall generate WIPO-compliant patent documents including the specification (title, abstract, description, drawings descriptions), claims, and figures list.
• FR-4.2: The system shall produce documents that expand on details rather than summarise, ensuring full disclosure as required by patent law.
• FR-4.3: The chatbot shall generate unambiguous, legally precise language suitable for patent claims, avoiding vague or overly broad phrasing that would fail examination.
• FR-4.4: The system shall support iterative document drafting, allowing users to review, edit, request revisions, and refine sections of the patent document through conversation.
• FR-4.5: The system shall generate properly numbered and formatted independent and dependent claims with correct claim dependencies.
• FR-4.6: The system shall produce documents compliant with USPTO, CIPO, and other major patent office formatting requirements.
• FR-4.7: The system shall support export of drafted documents in standard formats (PDF, DOCX) ready for filing.
• FR-4.8: The system shall maintain version history of all document drafts, enabling users to compare revisions and revert changes.

3.5 Claim Examination and Patentability Analysis

• FR-5.1: The chatbot shall act as a simulated patent examiner, analysing drafted claims for novelty against prior art retrieved from USPTO and Bing searches.
• FR-5.2: The chatbot shall assess claims for non-obviousness, providing reasoning about whether the invention represents a non-trivial advancement over existing art.
• FR-5.3: The chatbot shall check drafted documents for completeness and compliance, flagging missing sections, insufficient disclosure, or ambiguous claim language.
• FR-5.4: The system shall provide a structured patentability report summarising strengths, weaknesses, and recommendations for each claim.
• FR-5.5: The system shall perform a business utility analysis, helping users assess the commercial viability and market potential of their invention.

3.6 Team Collaboration and Discussion

• FR-6.1: Users shall be able to create named teams and invite other registered users by email or username.
• FR-6.2: The system shall provide a shared team workspace where all members can view and contribute to patent projects.
• FR-6.3: The system shall include a team discussion thread (chat) for each project, where members can discuss ideas, provide feedback, and debate patent strategies.
• FR-6.4: The chatbot shall be accessible within team discussions, allowing any member to invoke the bot for research, drafting assistance, or claim analysis within the team context.
• FR-6.5: The system shall implement role-based access control (RBAC) within teams, supporting roles such as Owner, Editor, and Viewer.
• FR-6.6: The system shall support co-editing of patent documents with real-time or near-real-time synchronisation and conflict resolution.
• FR-6.7: The system shall maintain an audit trail of all team member contributions and changes to documents.
• FR-6.8: Team discussions and AI-generated content shall be clearly attributed to distinguish between human and AI contributions.

3.7 Automated USPTO Filing

• FR-7.1: The system shall provide an automated filing feature that submits the finalised patent application to the USPTO electronically on behalf of the user, once the user has explicitly confirmed and authorised the submission.
• FR-7.2: Prior to automated filing, the system shall present the user with a comprehensive pre-filing review, displaying all documents, claims, drawings, and filing details for final approval.
• FR-7.3: The system shall require explicit multi-step user confirmation (review, confirm, and authenticate) before initiating any filing to prevent accidental submissions.
• FR-7.4: The system shall support integration with USPTO Patent Center for provisional and non-provisional patent applications.
• FR-7.5: The system shall guide users through the selection of the correct application type, entity size (micro, small, large), and fee calculation.
• FR-7.6: The system shall store and display filing confirmations, receipt numbers, and acknowledgements received from USPTO.
• FR-7.7: The system shall provide status tracking for filed applications, alerting users to any office actions, deadlines, or required responses.
• FR-7.8: The automated filing feature shall require users to provide their own USPTO credentials; the system shall never store credentials in plain text and shall use secure, encrypted storage with user-managed revocation.

3.8 Trademark Assistance and Multi-Jurisdiction Support

• FR-8.1: The chatbot shall assist users with trademark name creation, including availability checking and conflict analysis.
• FR-8.2: The system shall provide guidance for patent and trademark filing across multiple jurisdictions, including USPTO (USA), CIPO (Canada), and other major international patent offices.
• FR-8.3: The system shall generate jurisdiction-specific document formatting where requirements differ between patent offices.

3.9 Pro Tier: Multi-Model Verification

• FR-9.1: The Pro tier shall employ a multi-model debate architecture where the primary locally deployed LLM’s outputs are cross-verified by external models (e.g., Grok and Claude) via API calls.
• FR-9.2: The system shall present the synthesised results of the multi-model debate to the user, highlighting areas of agreement and disagreement between models.
• FR-9.3: The multi-model debate shall be applied to claim examination, prior-art analysis, and document review to increase accuracy and reduce hallucination.
• FR-9.4: The system shall clearly disclose to Pro tier users that their queries are processed by multiple AI providers and explain the data-sharing implications.

4. Non-Functional Requirements

4.1 Performance

• NFR-1.1: The chatbot shall respond to user messages within 5 seconds for standard queries and within 15 seconds for queries involving external API calls (Bing, USPTO).
• NFR-1.2: The system shall support a minimum of 2,000 concurrent users on the prototype GPU cluster without degradation of response quality.
• NFR-1.3: Document generation (full patent draft) shall complete within 60 seconds for standard-length applications.

4.2 Scalability

• NFR-2.1: The architecture shall support horizontal scaling from 4 GPUs (prototype) to 35+ GPUs (full deployment) without requiring application re-architecture.
• NFR-2.2: The system shall support a multi-site load balancer for distributing inference requests across geographically distributed GPU clusters in future phases.
• NFR-2.3: The system shall scale to support a target of 10,000 total subscribers within the first five years.

4.3 Security and Privacy

• NFR-3.1: All user data, invention descriptions, and patent documents shall be encrypted at rest (AES-256) and in transit (TLS 1.3).
• NFR-3.2: The system shall comply with COPPA (Children’s Online Privacy Protection Act) requirements for users under 13.
• NFR-3.3: Patent-related data shall be treated as trade secrets; the system shall enforce strict data isolation between users and teams.
• NFR-3.4: User invention data shall never be used for model training or shared with third parties without explicit consent.
• NFR-3.5: The system shall implement multi-factor authentication (MFA) for accounts and require additional authentication before automated filing actions.
• NFR-3.6: USPTO credentials shall be stored using industry-standard secrets management with user-controlled revocation.

4.4 Reliability and Availability

• NFR-4.1: The system shall target 99.9% uptime for the chatbot service.
• NFR-4.2: The system shall implement automatic failover from on-premise GPU cluster to cloud backup in the event of hardware failure.
• NFR-4.3: All user data and patent documents shall be backed up with a Recovery Point Objective (RPO) of 1 hour and a Recovery Time Objective (RTO) of 4 hours.

4.5 Usability

• NFR-5.1: The user interface shall be designed to be accessible to users as young as 10 years old, with clear navigation, age-appropriate language options, and interactive guidance.
• NFR-5.2: The system shall comply with WCAG 2.1 Level AA accessibility guidelines.
• NFR-5.3: The system shall support internationalisation (i18n) to enable future multi-language support.
• NFR-5.4: The onboarding experience shall include an interactive tutorial that walks new users through the patent process using the chatbot.

4.6 Compliance

• NFR-6.1: Generated patent documents shall conform to USPTO, CIPO, and WIPO formatting and content requirements.
• NFR-6.2: The system shall include disclaimers that AI-generated documents should be reviewed by a qualified patent attorney before filing, especially for high-value inventions.
• NFR-6.3: The system shall comply with applicable data protection regulations including GDPR for European users and CCPA for California users.

5. AI Model and Infrastructure Requirements

5.1 Model Selection

• MR-1.1: The final model shall be selected based on comparative evaluation across patent-specific benchmarks including document completeness, claim accuracy, legal language precision, and hallucination rate.
• MR-1.2: Initial model testing shall be conducted via OpenRouter to evaluate multiple models in the cloud before committing to a local deployment.

5.2 Fine-Tuning Approach

• MR-2.1: The selected model shall be fine-tuned using Low-Rank Adaptation (LoRA) to specialise it for patent document generation without catastrophic forgetting of general capabilities.
• MR-2.2: Fine-tuning datasets shall include high-quality patent documents from USPTO and WIPO, covering diverse technology domains and claim structures.
• MR-2.3: The fine-tuning process shall specifically target the model’s tendency to summarise, training it instead to expand and elaborate on technical details.
• MR-2.4: Fine-tuning shall be performed on the on-premise GPU cluster once models are selected and validated in the cloud.
• MR-2.5: The system shall support incremental fine-tuning with new patent data as it becomes available, without requiring full retraining.

5.3 Inference Infrastructure

• MR-3.1: Primary inference shall be served from the on-premise NVIDIA B300 GPU cluster (4 GPUs) with tensor-parallel configuration for optimal throughput.
• MR-3.2: Cost-effective inference for standard queries shall be offloaded to Mac Mini nodes with unified memory (64 GB RAM) where appropriate.
• MR-3.3: The system shall implement intelligent routing to direct requests to GPU or Mac Mini nodes based on query complexity and current load.
• MR-3.4: GPU thermals shall be continuously monitored with automated alerting and throttling to prevent hardware failure.
• MR-3.5: UPS (Uninterruptible Power Supply) protection shall be required for all inference hardware.

5.4 Cloud Backup and Redundancy

• MR-4.1: A cloud-hosted instance of the inference model shall be maintained as a backup for failover scenarios.
• MR-4.2: The system shall automatically route requests to the cloud backup when on-premise infrastructure is unavailable or overloaded.

6. External Interface Requirements

6.1 Bing Search API Integration

• EI-1.1: The system shall integrate with Microsoft Bing Web Search API for real-time topic research, retrieving relevant web pages, academic papers, and technical articles related to the user’s invention domain.
• EI-1.2: Search queries shall be constructed programmatically by the chatbot based on the user’s invention description, key concepts, and identified technology areas.
• EI-1.3: The system shall implement result caching and rate-limiting to optimise API usage and costs.

6.2 USPTO Integration

• EI-2.1: The system shall integrate with USPTO Patent Full-Text Database (PatFT) and Application Full-Text Database (AppFT) for prior-art searching.
• EI-2.2: The system shall integrate with USPTO Patent Center for automated electronic filing of provisional and non-provisional patent applications.
• EI-2.3: The system shall retrieve and parse USPTO fee schedules to calculate filing fees based on entity type and application category.
• EI-2.4: The system shall handle USPTO session management, form population, document attachment, and submission confirmation programmatically.

6.3 Third-Party LLM APIs (Pro Tier)

• EI-3.1: The Pro tier shall integrate with external LLM APIs (e.g., xAI Grok API, Anthropic Claude API) for the multi-model debate and verification feature.
• EI-3.2: API integration shall be modular, allowing new models to be added or swapped without significant refactoring.
• EI-3.3: The system shall handle API rate limits, errors, and timeouts gracefully, with fallback to single-model operation if external APIs are unavailable.

6.4 Payment Gateway

• EI-4.1: The system shall integrate with a PCI-DSS compliant payment processor (e.g., Stripe) for subscription billing.
• EI-4.2: The system shall support credit/debit card, Apple Pay, and Google Pay payment methods.

7. User Interface Requirements

• UI-1.1: The primary interface shall be a chat-based conversational UI with a clean, modern design suitable for all age groups.
• UI-1.2: The interface shall include a side panel for document preview, enabling users to view the evolving patent document alongside the chat conversation.
• UI-1.3: The system shall provide a dedicated project dashboard listing all active patent projects, their status, and quick-access actions.
• UI-1.4: The research results view shall display web search and prior-art results in an organised, filterable format with relevance indicators.
• UI-1.5: The team collaboration view shall include a shared chat thread, document editor, and member management panel.
• UI-1.6: The filing workflow shall present a step-by-step wizard with progress indicators, document previews, and confirmation gates.
• UI-1.7: The interface shall include a youth-friendly mode with simplified navigation, tooltips, gamification elements (progress badges, invention milestones), and educational pop-ups.
• UI-1.8: The system shall provide a rich-text document editor for manual editing of AI-generated patent documents, supporting headings, numbered paragraphs, inline figures, and claim formatting.

8. Subscription Tier Feature Matrix

9. System Architecture Overview

9.1 Frontend Layer

• Responsive web application built with a modern JavaScript framework (e.g., React or Next.js).
• Real-time WebSocket connection for streaming chatbot responses.
• Rich document editor component for patent document viewing and editing.
• Team chat and collaboration module with presence indicators.

9.2 Backend Layer

• RESTful API server handling authentication, project management, team operations, and filing workflows.
• WebSocket server for real-time chat and collaboration features.
• Background job queue for asynchronous tasks: document generation, USPTO filing, research aggregation, and fine-tuning jobs.
• Secure credential vault for USPTO authentication tokens.

9.3 AI / ML Layer

• Locally deployed open-source LLM with LoRA fine-tuning adapters, served via an inference framework (e.g., vLLM or TGI).
• RAG pipeline connecting Bing Search results and USPTO prior-art data to the LLM context.
• Mac Mini inference nodes for lightweight query routing.
• Cloud-hosted model replica for failover and overflow.
• Pro tier orchestration layer for multi-model debate (local LLM + Grok API + Claude API).

9.4 Data Layer

• Relational database (e.g., PostgreSQL) for user accounts, projects, teams, and filing records.
• Document storage (e.g., S3-compatible object store) for patent documents, drafts, and exports.
• Vector database (e.g., Pinecone, Weaviate, or pgvector) for semantic search over patent corpus and prior-art data.
• Redis or equivalent for session management, caching, and rate limiting.

10. Constraints and Limitations

• C-1: The chatbot is not a licensed patent attorney and all generated documents carry a disclaimer recommending professional legal review before filing.
• C-2: Automated USPTO filing depends on the continued availability and stability of USPTO’s electronic filing interfaces. Changes to USPTO systems may require adaptation.
• C-3: The quality of patent documents is bounded by the capabilities of the underlying LLM and the quality of fine-tuning data. Ongoing model evaluation and fine-tuning iterations will be required.
• C-4: Multi-model debate in the Pro tier depends on the availability and pricing of third-party LLM APIs (Grok, Claude), which are outside the system’s control.
• C-5: Prior-art search via Bing and USPTO may not be exhaustive. The system shall clearly communicate that no prior-art search tool is a substitute for a professional patentability opinion.
• C-6: Youth-focused features must comply with COPPA and similar regulations, which may restrict certain data collection and feature availability for minors.

11. Future Enhancements

The following items are identified as potential future features beyond the initial prototype scope:

• AI-powered patent drawing generation using image generation models.
• Integration with additional patent offices beyond USPTO and CIPO (EPO, JPO, KIPO, etc.).
• Native mobile applications for iOS and Android.
• AI-driven patent portfolio management and analytics dashboard.
• Marketplace connecting inventors with patent attorneys for professional review.
• Integration with university and school platforms for educational programs.
• Automated response drafting for USPTO office actions.
• Multi-site GPU load balancing across distributed Cloud Wash facilities.
• Patent valuation and licensing recommendation engine.

12. Appendix

12.1 Requirement Traceability Matrix

A detailed traceability matrix mapping each requirement ID to design components, test cases, and acceptance criteria will be maintained as a separate living document and updated throughout the development lifecycle.

12.2 Revision History