ResearchFlow: AI-Powered Research Assistant

Inspiration

As research beginners ourselves, we've experienced the frustration of new professional terms, scattered notes, forgotten experiment connections, and the overwhelming task of literature review. We watched brilliant PhD students lose track of their own insights and struggle to see how their experiments build upon each other. We envisioned an AI partner that could transform chaotic research processes into clear, connected knowledge graphs - making scientific discovery more intuitive and collaborative.

What it does

ResearchFlow is an intelligent research companion that revolutionizes how scientists manage their experimental work:

Conversational Experiment Planning: Chat with AI about research interests and receive tailored experiment suggestions with detailed motivations and expected outcomes
Automatic Knowledge Graphs: Visualizes experiments as connected nodes, showing relationships like "leads_to," "supports," or "refutes" between studies
Smart Literature Integration: AI automatically finds and suggests relevant papers for each experiment, with relationship mapping (prior work, builds upon, contrasts)
Visual Progress Tracking: Color-coded nodes (green for completed, yellow for planned, red for postponed) provide instant project status overview
Team Collaboration: Shared dashboards, meeting notes, and professor feedback integration for seamless lab communication

How we built it

Full-Stack TypeScript Architecture:

Frontend: React 18 + TypeScript with ReactFlow for graph visualization, Ant Design for UI components, and Tailwind CSS for styling
Backend: Dual-service architecture with FastAPI (Python) for data management and Express.js for AI integration
AI Integration: OpenAI GPT-4 for conversation and suggestions, OpenAlex API for literature search, Gemini for document analysis
Database: SQLAlchemy ORM with SQLite for experiments, relationships, and literature references
Real-time Features: WebSocket connections for live collaboration and Server-Sent Events for feedback updates

Key Technical Innovations:

Dynamic graph layout using Dagre algorithm for optimal node positioning
Template-based AI prompt engineering for consistent, high-quality suggestions
Modular tool system allowing AI to call specific functions for graph manipulation
Caching layer for literature suggestions to improve response times

Challenges we ran into

AI Consistency: Creating reliable prompts that generate structured experiment suggestions while maintaining conversational flow
Graph Visualization: Balancing automatic layout algorithms with user interaction needs - making complex research networks readable
Context Management: Maintaining conversation context across multiple AI interactions while preserving experiment relationships
Literature Integration: Handling unreliable external APIs and implementing fallback strategies for paper validation
Real-time Synchronization: Ensuring graph updates propagate correctly across different interface components

Accomplishments that we're proud of

Complete Working System: Built a full-stack application with working AI integration in record time
Intelligent Auto-connections: AI automatically creates logical relationships between experiments without user intervention
Professional UI/UX: Implemented enterprise-grade interface with smooth animations and intuitive interactions
Scalable Architecture: Designed modular backend that supports multiple AI models and can handle growing research teams
Real Research Value: Created something we'd actually use for our own research projects

What we learned

AI Integration Complexity: Learned that reliable AI systems require extensive error handling and fallback strategies
Graph Visualization Challenges: Discovered that automatic layout algorithms need careful tuning for different research domains
User-Centric Design: Realized that even advanced AI features need simple, intuitive interfaces to be truly useful
Full-Stack Coordination: Gained experience managing complex state synchronization between multiple services and real-time updates
Research Workflow Understanding: Deepened our appreciation for how researchers actually work and the tools they need

What's next for ResearchFlow: AI-Powered Research Assistant

Immediate Enhancements:

Multi-modal Input: Support for uploaded papers, images, and data files to generate experiment suggestions
Advanced Analytics: Research trend analysis, gap identification, and impact prediction
Export Capabilities: Generate research proposals, progress reports, and presentation slides automatically

Long-term Vision:

Institutional Integration: Connect with university systems, grant databases, and institutional repositories
Collaborative Networks: Enable cross-laboratory collaboration and expertise matching
Predictive Research: AI that suggests breakthrough research directions based on current trends and gaps
Commercial Deployment: Transform into a SaaS platform serving research institutions globally

ResearchFlow represents the future of research management - where AI doesn't replace researchers, but empowers them to focus on discovery rather than organization.