PatentBot

An AI-powered system that ingests engineering/scientific literature, builds domain expertise, and identifies opportunities for novel patents by finding gaps, combinations, and unexplored intersections in the knowledge space.

Vision

Turn a book library into a patent factory.

Most patents come from combining existing ideas in novel ways. PatentBot:

1. Ingests technical literature (PDFs, papers, textbooks)
2. Builds structured knowledge graphs per domain
3. Identifies “white space” — unexplored combinations and gaps
4. Generates patent-worthy invention disclosures

Architecture

┌─────────────────────────────────────────────────────────────┐
│                     KNOWLEDGE LAYER                         │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐  │
│  │ Engineering  │    │   Physics/   │    │  Materials   │  │
│  │   Corpus     │    │   Quantum    │    │   Science    │  │
│  │  (1000+ PDFs)│    │   Corpus     │    │   Corpus     │  │
│  └──────┬───────┘    └──────┬───────┘    └──────┬───────┘  │
│         │                   │                   │          │
│         └───────────────────┼───────────────────┘          │
│                             ▼                              │
│                    ┌────────────────┐                      │
│                    │  RAG Pipeline  │                      │
│                    │  (Embeddings)  │                      │
│                    └────────┬───────┘                      │
│                             │                              │
└─────────────────────────────┼──────────────────────────────┘
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                     ANALYSIS LAYER                          │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐  │
│  │  Knowledge   │    │    Gap       │    │   Prior Art  │  │
│  │    Graph     │◄──►│  Detection   │◄──►│    Search    │  │
│  │  Extractor   │    │   Engine     │    │    Agent     │  │
│  └──────────────┘    └──────────────┘    └──────────────┘  │
│                             │                              │
└─────────────────────────────┼──────────────────────────────┘
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                     OUTPUT LAYER                            │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌──────────────────────────────────────────────────────┐  │
│  │              Patent Disclosure Generator              │  │
│  │                                                       │  │
│  │  • Title & Abstract                                   │  │
│  │  • Problem Statement                                  │  │
│  │  • Novel Solution                                     │  │
│  │  • Claims (Independent + Dependent)                   │  │
│  │  • Prior Art References                               │  │
│  │  • Figures/Diagrams (conceptual)                      │  │
│  └──────────────────────────────────────────────────────┘  │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Core Components

1. Corpus Ingestion

• Input: PDF/EPUB technical books, papers, patents
• Processing: OCR → chunking → embeddings
• Storage: Vector DB (local ChromaDB or Qdrant)
• Metadata: Domain tags, publication date, citation graph

2. Knowledge Graph Extraction

• Extract entities: materials, processes, properties, applications
• Map relationships: “X enables Y”, “A improves B”, “C replaces D”
• Build domain ontology automatically from corpus

3. Gap Detection Engine

• Combination gaps: A+B exists, B+C exists, but A+C doesn’t
• Property gaps: Material X has properties P,Q but no one’s tested R
• Application gaps: Technique T used in domain D1 but not D2
• Temporal gaps: Old idea + new capability = new invention

4. Prior Art Agent

• Search USPTO, Google Patents, arXiv
• Validate novelty before generating disclosure
• Find closest prior art for differentiation

5. Disclosure Generator

• LLM-powered patent drafting
• Structured output: claims, abstract, description
• Human review workflow

Sub-Projects

🏺 Pharmakon Mining

Cross-axis patent discovery from Ancient Greek pharmacology. Mining 2,500-year-old pharmaceutical knowledge (theriac, PGM, Dioscorides) and mapping to modern science for novel patentable formulations.

Target Domains (Phase 1)

Tech Stack

Development Phases

Phase 1: Foundation (MVP)

• Set up RAG pipeline with engineering corpus
• Basic entity extraction (materials, processes)
• Simple gap detection: “What combinations don’t exist?”
• Manual patent drafting from insights

Phase 2: Intelligence

• Knowledge graph construction
• Prior art search integration
• Automated novelty scoring
• Disclosure template generation

Phase 3: Automation

• Full disclosure generator
• Patent attorney review workflow
• Filing assistance
• Portfolio management

Success Metrics

• Disclosures generated: Target 10+/month after Phase 2
• Novelty rate: >50% pass prior art check
• Filing rate: 1-2 provisional patents/quarter
• Time to disclosure: <4 hours from gap identification

Revenue Model

1. Own patents — File provisional, license or sell
2. Service — Generate disclosures for clients ($500-2000/disclosure)
3. SaaS — Patent discovery platform for R&D teams

Risks & Mitigations

Related Projects

• Knowledge RAG System — Foundation
• Book Library — Source corpus
• NTS — Potential service offering

Next Steps

1. Build engineering corpus in RAG system
2. Test entity extraction on 10 sample books
3. Prototype gap detection algorithm
4. Generate first 5 invention concepts manually
5. Validate novelty with prior art search

“The best way to predict the future is to invent it.” — Alan Kay