Hook: Turn “already patented” ideas into clear, actionable innovation pathways.
- 🎬 Demo :
🚧 This is a prototype system
There are:
Known defects, Incomplete integrations & Areas needing optimization
However, this project is an excellent end-to-end learning system.
Every founder, researcher, or innovator hits the same wall:
“I found a similar idea patented/productized already….. so what now?”
Traditional patent tools tell you:
- What exists ❌
- Who owns it ❌
But they don’t tell you how to move forward.
You are stuck between:
- Reinventing something already patented ❌
- Risking infringement ❌
- Or abandoning your idea ❌
This project introduces:
🎯 Innovation Delta = The specific technical gap that makes your idea patent-worthy
Instead of stopping at search results, this system:
- Breaks your idea into features
- Maps them to prior art
- Identifies idea saturation vs novelty
- Suggests how to differentiate
This is not just another RAG system.
It performs:
- Feature-level semantic decomposition (not just document retrieval)
- Evidence-backed overlap detection across prior art
- Multi-step Agentic RAG via MCP (Microservice Command Protocol) — orchestrating retrieval, evidence mapping, and novelty scoring as specialized agents
- 🚀 Difference Engineering: converts overlap into actionable innovation pathways
It doesn’t just explain what exists — it tells you how to make your idea distinct and patent-worthy.
| Layer | Technology |
|---|---|
| Frontend | Next.js |
| Runtime | Node.js |
| Backend | FastAPI |
| LLM | Llama 3.2 |
| Embeddings | Jina Embeddings |
| Database | PostgreSQL |
| Patent Data | Lens.org API |
| Orchestration | MCP (Microservice Command Protocol) |
flowchart TD
A["User Input Idea"] --> B["Frontend - Next.js"]
B --> C["Backend Orchestrator - FastAPI"]
C --> D1["MCP Retrieval Service"]
D1 --> D2["Lens.org API"]
D1 --> D3["Google Patents Fallback"]
D2 --> E["Patent Results"]
D3 --> E
E --> F["Jina Embeddings Vectorization"]
F --> G["MCP Evidence Service"]
G --> H["Feature-to-Passage Mapping"]
H --> I["MCP Novelty Service"]
I --> J["Overlap + Saturation Analysis"]
J --> K["LLM - Llama 3.2"]
K --> L["Innovation Delta Generation"]
L --> M["MCP Report Service"]
M --> N["Structured Innovation Report"]
N --> O["Frontend Visualization"]
You must obtain Lens.org API access for live patent retrieval:
👉 https://www.lens.org/lens/user/subscriptions
run_all.bat (run the batch file in command prompt/powershell)
Access: Open browser: http://127.0.0.1:5006/idea-input
🧪 How to Use
1️⃣ Submit Concept
Enter your idea in natural language
2️⃣ Semantic Analysis
System performs: Patent retrieval, Feature extraction, Evidence mapping
3️⃣ Review Innovation Report
You get: Feature decomposition, Prior art mapping, Novelty map, CPC codes
🚀 Innovation Deltas 📌 Example Scenario
🧾 Input
Idea title: AI Food Expiry Tracker Domain: AI, computer-vision, smart-home Problem statement: Households waste food because people forget what's in their fridge and when it expires. A phone camera scans fridge contents daily, identifies items using computer vision, estimates expiry dates, and sends timely alerts to use ingredients before they spoil. Objectives: Reduce food waste, expiry alerts, recipe suggestions from near-expiry items Constraints: Works with standard phone camera, no smart fridge required, offline inference Tags: computer-vision, food-waste, edge-AI, smart-home
📊 Output
Existing Coverage & 🚀 Innovation Delta Suggestions:
- Emphasize ai food expiry tracker as the likely differentiator.
- Describe implementation constraints for ai food expiry tracker more concretely.
🎯 Result: Not a search result, but a patentable direction
Some UI sections may not populate if Lens API lacks metadata 📄 PDF export format is incorrect Occasional fallback to demo data if live retrieval fails
- Multi-Source Synthesis Add Google Patents + ArXiv Improve recall and coverage
- Automated Claim Drafting Generate initial patent claims Based on Innovation Deltas
- Interactive Patent Landscape 2D / 3D visualization Identify white-space innovation zones
This project is designed as a hands-on system to understand modern AI architecture.
🧠 What You’ll Learn
- How Agentic RAG differs from standard RAG
- How MCP enables modular AI pipelines
- How embeddings power semantic retrieval
- How prompting drives structured innovation
🔪 Core Concepts Explained
- Agentic RAG
Instead of: Retrieve → Answer, We do: Retrieve → Compare → Analyze → Suggest
- MCP (Microservice Command Protocol)
Each capability is a separate service: Service Role, Retrieval, Patent search, Evidence, Mapping features, Novelty, Overlap detection, Report & Output generation
👉 Orchestrator = AI system coordinator
- Feature Decomposition
Input idea → structured features:
[ "Camera-based monitoring", "ERP integration", "Prediction model", "Temporal lag detection" ] 4. Innovation Delta Prompting
The LLM is guided to:
- Identify saturation zones
- Detect gaps
- Suggest differentiation 🤝 Contributing
This is a learning + innovation project.
If you're interested in:
AI systems, Patent intelligence, Agentic workflows