PatentSight

✨ Overview

Patent Innovation Predictor is an intelligent multi-agent system designed to automate the analysis, retrieval, and forecasting of patent innovations. Built with CrewAI, LangChain, Ollama, and OpenSearch, the system enables powerful Retrieval-Augmented Generation (RAG) workflows tailored for patent data. It dynamically analyzes any research domain (e.g., lithium batteries, hydrogen storage, quantum computing) to uncover trends and predict future innovations.

📊 Key Features

• ✨ Agentic Workflow: Uses four CrewAI agents with well-defined roles.
• 🔮 Retrieval-Augmented Generation (RAG): Real-time access to OpenSearch-powered patent database.
• 📆 Custom Research Domains: Works with any field the user inputs at runtime.
• 🧑‍💼 LLM Reasoning with Data: Merges patent metadata with LLM-driven insights.
• ⚡ Ollama for Local LLMs: Ensures privacy, speed, and no external dependencies.
• 🔹 Modular Design: Easily extensible with new tools, agents, or data connectors.

---

🛠️ Tech Stack

Component	Technology	Purpose
Language	Python 3.9+	Core logic and orchestration
Agents	CrewAI	Multi-agent workflow manager
Prompt/LLM Chains	LangChain	Prompt templates, chaining, and LLM abstraction
LLM Server	Ollama + LangChain	Local LLM execution (LLaMA3, Mistral, etc.)
Search Engine	OpenSearch	Patent indexing, keyword & vector search
Embeddings	nomic-embed-text	Patent vector embeddings via Ollama
Data API	SerpAPI	Real-time patent data scraping
Containerization	Docker	Runs Ollama & OpenSearch locally
Storage	JSON + OpenSearch	Indexed patents and search results

---

🪤 Agents & Tasks

Agent	Purpose	Tools Used
💼 Research Director	Define research goals, key focus areas, time ranges	LLM only
🧰 Patent Retriever	Query OpenSearch, retrieve & organize patents	search_patents, search_by_date_range
📊 Data Analyst	Detect trends, tech evolution, and innovation hotspots	analyze_patent_trends
🌌 Innovation Forecaster	Predict future breakthroughs and recommend R&D areas	LLM only

---

🛩️ How It Works

1. Patent Collection:

   • Run information_collector.py to fetch patent data from SerpAPI.
   • Save as JSON and load into OpenSearch using opensearch_client.py.

2. Embedding Indexing:

   • Generate vector embeddings using embedding.py via nomic-embed-text.
   • Store embeddings in OpenSearch for hybrid (keyword + vector) search.

3. Main Pipeline Execution:

   • Execute agentic_rag.py.
   • Prompts user for research area (e.g., "Hydrogen Storage") and LLM model (e.g., "llama3").
   • Checks if Ollama and OpenSearch are running.
   • Instantiates agents and their tools.
   • Agents sequentially execute tasks, passing data forward.
   • Final report is saved to /results/.

---

⚙️ Setup Guide

1. Clone the Repository

git clone <repo-url>

2. Set Up Virtual Environment

python -m venv .venv
.\.venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Environment Variables

Create a .env file with your SerpAPI key:

SERPAPI_API_KEY=your_key_here

5. Start Services

OpenSearch

Follow instructions from OpenSearch Docs Ensure it's running at localhost:9200.

Ollama

docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
docker exec -it ollama ollama pull llama3

For embeddings:

docker exec -it ollama ollama pull nomic-embed-text

6. Index Patent Data

python information_collector.py  # collect
python opensearch_client.py      # index into OpenSearch
python embedding.py              # add embeddings

7. Run Main Pipeline

python agentic_rag.py

---

🗂️ File Structure

Ai_Agent/
├── agentic_rag.py           # Main CLI pipeline
├── patent_crew.py           # Agents and tasks setup
├── opensearch_client.py     # Index management
├── embedding.py             # Embedding generator
├── information_collector.py # Patent scraper (SerpAPI)
├── helper.py                # Utility functions
├── .env                     # Secrets (API keys)
├── requirements.txt         # Python dependencies
├── files/                   # Raw patent data
├── results/                 # Analysis output
└── README.md                # Project documentation

---

⚡ Advanced Usage & Customization

• Switch LLM Model: At runtime, specify any model available in Ollama (e.g., "mistral").
• Change Research Domain: Just input a new research field at runtime.
• Add New Agents: Extend patent_crew.py to include critics, validators, or enrichment agents.
• Output Formats: Change agent outputs to JSON/CSV for dashboards or API use.
• Web Interface: Integrate with FastAPI or Streamlit for interactive UI.

---

🔎 Example

Enter the research area to analyze (default: Lithium Battery): Hydrogen Storage
Enter the Ollama model to use (default: llama2): llama3

[...Agents running...]

Analysis completed and saved to: /results/patent_analysis_20250704_153000.txt

---

🚀 Future Improvements

• Add Agent Memory for long-term context.
• Introduce Critic Agent for validation.
• Integrate visual analytics (charts/timelines).
• Expose as API or Web App.

---

Pipeline of Patent Innovation Predictor

1. User Input

   • User starts the CLI (python agentic_rag.py).
   • User enters the research area (e.g., "Lithium Battery") and selects the Ollama LLM model (e.g., "llama3").

   

2. Environment & Service Checks

   • Script checks if Ollama (LLM server) and OpenSearch (vector DB) are running and accessible.
   • Ensures required models are available in Ollama.

3. Patent Data Collection & Indexing

   • information_collector.py fetches patent data from SerpAPI for the specified research area.
   • Data is saved as JSON files in the files/ or results/ directory.
   • embedding.py generates vector embeddings for patent abstracts using Ollama’s embedding model.
   • opensearch_client.py creates the OpenSearch index (if not present) and ingests the patent data with embeddings.

4. Agentic RAG Workflow (CrewAI Orchestration)

   • Four agents are instantiated:
     • Research Director: Defines the research plan for the chosen area.
     • Patent Retriever: Uses OpenSearch tools to retrieve and organize relevant patents.
     • Data Analyst: Analyzes trends, patterns, and company focus from the retrieved data.
     • Innovation Forecaster: Predicts future breakthroughs and R&D priorities.
   • Each agent executes its task in sequence, passing outputs to the next agent.

5. Patent Search & Retrieval

   • Agents use CrewAI tools to query OpenSearch for patents matching the research area and time window.
   • Results are grouped, summarized, and prepared for analysis.

   

6. Trend Analysis & Forecasting

   • Data Analyst agent identifies innovation trends, key companies, and emerging technologies.
   • Innovation Forecaster agent predicts future directions, R&D priorities, and disruptive trends.

7. Reporting

   • The final output (comprehensive analysis and forecast) is saved to a timestamped .txt file in the project directory.

8. User Review

   • User reviews the saved report for insights, trends, and recommendations.

   

---

Technologies Involved at Each Step:

• Python: Orchestration, agent logic, and CLI.
• CrewAI: Multi-agent workflow and task management.
• LangChain: LLM prompt management and chaining.
• Ollama: Local LLM and embedding model serving (via Docker).
• OpenSearch: Patent data storage, keyword and vector search.
• SerpAPI: Patent data collection from Google Patents.
• Docker: Containerization for Ollama and OpenSearch services.

---

Summary Diagram:

User Input
   ↓
Service Checks (Ollama, OpenSearch)
   ↓
Patent Data Collection (SerpAPI → JSON)
   ↓
Embedding Generation (Ollama)
   ↓
Indexing in OpenSearch
   ↓
CrewAI Agentic Pipeline:
   Research Plan → Patent Retrieval → Trend Analysis → Innovation Forecast
   ↓
Report Generation (.txt)
   ↓
User Review#