A newer version of the Streamlit SDK is available:
1.53.1
marp: true
theme: default
paginate: true
backgroundColor: null
backgroundImage: url('https://marp.app/assets/hero-background.svg')
SPARKNET
AI-Powered Patent Valorization System
A Multi-Agent Platform for Technology Transfer
Hamdan November 2025
System Architecture & Components
βββββββββββββββββββββ SPARKNET Platform βββββββββββββββββββββββββ
β β
β Frontend (Next.js) ββββββΊ Backend (FastAPI + LangGraph) β
β Port 3001 Port 8001 β
β β β
β ββββββββββββββββββββΌββββββββββββββ β
β β LangGraph State Machine β β
β β Workflow Orchestrator β β
β ββββββββββββββββ¬ββββββββββββββββββ β
β β β
β ββββββββ STARTUP AGENTS (4) βββ΄ββββββββββββββββββββββ β
β β β β
β β ββββββββββββ ββββββββββββ ββββββββββββ ββββββββββ β
β β β Planner β β Critic β β Memory β β Vision β β
β β β Agent β β Agent β β Agent β β OCR β β
β β βqwen2.5 β β mistral β β ChromaDB β βllava:7bβ β
β β β :14b β β :latest β β Vector β β β β
β β ββββββββββββ ββββββββββββ ββββββββββββ ββββββββββ β
β βββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
β β
β βββββ RUNTIME AGENTS (4) - Created per workflow βββββ β
β β β β
β β ββββββββββββ ββββββββββββ ββββββββββββ ββββββββββββ β
β β βDocument β β Market β βMatching β βOutreach β β
β β βAnalysis β β Analysis β β Agent β β Agent β β
β β βllama3.1 β βllama3.1 β βllama3.1 β βllama3.1 β β
β β β :8b β β :8b β β :8b β β :8b β β
β β ββββββββββββ ββββββββββββ ββββββββββββ ββββββββββββ β
β βββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
Key Components: 8 Agents β’ 4 LLM Models β’ State Machine β’ Vector Store
Functional Workflow: Patent Wake-Up Pipeline
Phase 1: Orchestration π―
- PlannerAgent (qwen2.5:14b): Decomposes task into executable subtasks
- MemoryAgent (ChromaDB): Retrieves relevant context from past analyses
- LangGraph routes workflow to Patent Wake-Up scenario
Phase 2: Sequential Analysis (4-Step Pipeline) π€
Step 1: Document Analysis π
- DocumentAnalysisAgent (llama3.1:8b) + VisionOCRAgent (llava:7b)
- Extracts text using PyMuPDF, processes images with OCR
- Identifies: Title, Abstract, Claims, Technical Domains, TRL Level
- Output: Patent Analysis Model with 1+ innovations
Step 2: Market Analysis π
- MarketAnalysisAgent (llama3.1:8b)
- Analyzes commercialization opportunities based on patent data
- Identifies market segments, competitive landscape
- Output: 4-5 Market Opportunities with sizing estimates
Step 3: Partner Matching π€
- MatchmakingAgent (llama3.1:8b)
- Queries MemoryAgent for stakeholder profiles from vector store
- Scores matches based on technology alignment
- Output: Top 10 potential partners ranked by compatibility
Step 4: Brief Creation π
- OutreachAgent (llama3.1:8b)
- Generates PDF valorization brief for stakeholder outreach
- Includes executive summary, technical details, business case
- Output: PDF document ready for distribution
Phase 3: Quality Validation β
- CriticAgent (mistral:latest): Validates output quality (threshold: 0.80)
- Stores successful episodes in MemoryAgent for future learning
- Returns results via WebSocket to frontend dashboard
Live Demonstration & Results
Example Analysis: Toyota Hydrogen Fuel Cell Initiative
| Metric | Result |
|---|---|
| Title | "Toyota Opens Door to Hydrogen Future" |
| Technical Domains | Automotive β’ Clean Energy β’ Fuel Cells |
| TRL Level | 8/9 (System Complete & Qualified) |
| Commercialization | HIGH |
| Key Innovations | β’ 5,680 patents royalty-free β’ High-pressure Hβ storage β’ Fuel cell stack optimization |
| Applications | Hydrogen vehicles β’ Power systems Industrial fuel cells |
System Status β
- Performance: Sub-2 minute analysis per document (117s avg)
- Accuracy: Multi-model validation with quality score β₯ 0.80
- Real-time Updates: WebSocket streaming for live progress
- Deployment:
- Frontend: http://172.24.50.21:3001
- Backend API: http://172.24.50.21:8001
Impact & Next Steps
Current Capabilities β
β Automated patent document analysis β Technology readiness assessment (TRL) β Multi-domain commercialization evaluation β Real-time web interface with workflow visualization
Value Proposition
Problem: Manual patent analysis takes days and requires domain experts Solution: SPARKNET reduces analysis time from days to < 1 minute Benefit: Universities can rapidly assess entire patent portfolios for licensing
Future Enhancements
- Batch processing for large patent portfolios
- Industry partner matching database
- Automated technology brief generation
- Integration with patent databases (USPTO, EPO)
Thank you!
Questions?
Live Demo URLs:
- Frontend: http://172.24.50.21:3001
- API Documentation: http://172.24.50.21:8001/api/docs
- API Health Check: http://172.24.50.21:8001/api/health