Networkify.ai: Supercharge your connections 🚀
AI-powered people search 🙍♂️🔍
Inspiration 💡
We created Networkify.ai after noticing a common problem: most professionals have hundreds of LinkedIn connections but rarely use them effectively. The average LinkedIn user has over 900 connections, yet studies show only about 2.5% are actively utilized for professional growth. We wanted to build a tool that helps people discover the hidden value in their professional networks.
Problem Statement 🎯
Professional networks represent immense untapped potential across multiple business domains:
For HR and Recruiting:
- Talent acquisition teams struggle to find qualified candidates through traditional channels
- Internal mobility is hindered by lack of visibility into employee skill networks
For Business Development:
- Sales teams miss opportunities for warm introductions to potential clients
- Partnership managers lack visibility into second-degree connections
- Relationship mapping between organizations is manual and incomplete
What it does 🔍
Networkify.ai is an AI-powered network analysis tool that helps professionals unlock the value in their LinkedIn connections:
- Visualizes your network as an interactive graph, showing connections, companies, and skills 📊
- Answers natural language questions about your network ("Who works at Google?" or "Who has data science skills?") 💬
- Identifies key influencers and strategic connection opportunities 🎯
- Discovers skill clusters to help you find experts for any project 🧩
- Compares multiple networks to find common connections and complementary strengths 🤝
How we built it 🛠️
- Exported LinkedIn data and processed the CSV files + generated synthetic data (like skills) to enrich the dataset
Technical Architecture
We built Networkify.ai using a sophisticated stack that combines graph database technology with AI:
- ArangoDB for efficient graph storage and querying
- NetworkX with NVIDIA cuGraph for GPU-accelerated graph algorithms
- LangChain and OpenAI for natural language processing
- Gradio for an intuitive user interface
Dynamic Query Routing
The core innovation of Networkify.ai is our intelligent query routing system that determines the optimal execution path based on query complexity:
def determine_query_type(query):
"""Uses LLM to determine the best query type for a given question"""
response = llm.invoke(f"""
I need to determine the best query type for this question: "{query}"
Classify into: "SIMPLE QUERY", "COMPLEX QUERY", or "HYBRID QUERY".
""").content
if "SIMPLE QUERY" in response:
return "Simple Query", text_to_aql_to_text
# ... routing logic for other query types
Hybrid Query Execution
For complex questions requiring both data retrieval and advanced analytics, our hybrid query execution system combines AQL and NetworkX/cuGraph:
def hybrid_query_execution(query):
# Step 1: Generate and execute AQL to retrieve relevant data subset
aql_response = llm.invoke(f"Write AQL query for: {query}")
cursor = db.aql.execute(aql_response)
results = [doc for doc in cursor]
# Step 2: Select and apply appropriate graph algorithm
algorithm = llm.invoke(f"Which algorithm for: {query}").strip().lower()
subgraph = create_subgraph_from_results(results)
if 'pagerank' in algorithm:
metrics = nx.pagerank(subgraph)
# ... other algorithm options
This hybrid approach enables sophisticated queries like:
- "Who are the most influential people at Microsoft?"
- "Find potential mentors in data science who are well-connected to hiring managers"
- "Identify the best person to introduce me to decision-makers at Google"
Multi-dimensional Graph Creation
We create a rich graph structure with three types of relationships:
Create company-based edges
for company in companies:
same_company = combined_df[combined_df['Company'] == company]['id'].tolist()
# Connect people at same company (limited to maintain realistic density)
# ...
Create skill-based edges
for skill, people in skill_to_people.items():
# Connect people with shared skills
# ...
Challenges we ran into 🧗
- Query Classification Accuracy: Ensuring the LLM correctly identifies query types required careful prompt engineering and testing with diverse query patterns.
- Graph Algorithm Selection: Automatically determining which graph algorithm is most appropriate for a given query was challenging. We implemented a context-aware selection system that considers both the query intent and the data structure.
- Performance Optimization: GPU-accelerated graph algorithms required careful memory management, especially for large networks. We implemented selective subgraph extraction to process only relevant portions of the graph.
- Realistic Network Generation: Creating a realistic professional network from limited data required sophisticated edge creation strategies that balance density and meaningful connections.
Accomplishments that we're proud of 🏆
- Intelligent Query Routing: Our system automatically determines whether a question needs simple data retrieval (AQL), complex graph algorithms (NetworkX/cuGraph), or both (hybrid query).
- GPU-Accelerated Analytics: By integrating cuGraph, we've enabled high-performance graph algorithms that can process large networks much faster than CPU-based approaches.
- Natural Language Graph Querying: We've created an intuitive interface that translates natural language questions into precise graph operations, making complex network analysis accessible to non-technical users.
- Multi-dimensional Relationship Analysis: Our system captures and analyzes different types of professional relationships (company-based, skill-based, network-based), providing richer insights than traditional connection models.
What's next for Networkify.ai 🔮
- Direct LinkedIn Integration: Automating the data import process through LinkedIn's API
- Temporal Analysis: Tracking how networks evolve over time to identify emerging trends and opportunities
- Recommendation Engine: Suggesting new connections that would strategically enhance your network
- Team Collaboration: Allowing teams to combine their networks for more comprehensive analysis
- Industry-Specific Insights: Tailored analytics for sales teams, recruiters, and entrepreneurs
Log in or sign up for Devpost to join the conversation.