​
Introduction

We want to create a Python script that uses HelixDB to help students find professors based on their research areas and bio. For example a user could ask
  • “What professors does High-Energy Physics research?”
  • “What professors are working in the X University?”
  • “What professors are working in the Computer Science department?”
  • “What professors are working in the X University and are working in the Computer Science department?”
  • “I like doing research in Large Language Models, can you recommend me some professors doing this in X University?”

​
The dataset

In this example, we have a dataset on Professors with fields like:
  • Name
  • Title
  • Department(s)
  • University
  • Their page URL
  • Short biography
  • Key Research Areas and their descriptions
  • Lab(s), the name, lab’s research focus
We will be ingesting this data from a JSON file, an example is shown below: 
{
    "name": "James",
    "title": "Assistant Professor",
    "page": "https://james.com",
    "department": ["Computer Science"],
    "university": ["Uni X"],
    "bio": "James is an Assistant Professor whose work sits at the intersection
     of basketball analytics, computer vision, and large-scale machine learning. 
     His research focuses on turning raw player-tracking video, wearable-sensor streams,
     and play-by-play logs into actionable insights for teams, coaches, and broadcasters. 
     Signature projects include ShotNet— a deep learning model that predicts shot success
     probability in real time— and DunkGPT, a language model fine-tuned on millions of play descriptions
     to generate advanced scouting reports.",
    "key_research_areas": [
        {
            "area": "Computer Vision for Basketball",
            "description": "Designing CNN and Transformer architectures that track
             player pose, ball trajectory, and court zones to quantify defensive pressure and shooting mechanics."
        },
        {
            "area": "Predictive Modelling & Simulation",
            "description": "Building Monte-Carlo and sequence models that forecast 
            possession outcomes and season performance using play-by-play and spatial data." 
        },
        {
            "area": "Sports Analytics with Large Language Models",
            "description": "Leveraging LLMs to explain model outputs, auto-generate commentary,
             and mine historical game archives for strategic patterns." 
        },
        {
            "area": "Wearable Sensor Data Mining",
            "description": "Applying time-series and graph learning techniques to
             inertial-measurement signals for fatigue monitoring and injury prevention." 
        },
        {
            "area": "Fairness & Ethics in Sports AI",
            "description": "Studying algorithmic bias and ensuring equitable analytics across different leagues,
             genders, and play styles." 
        }
    ],
    "labs": [
        {
            "name": "Basketball Data Science Lab",
            "research_focus": "An interdisciplinary group combining data science, biomechanics, 
            and sport psychology to create next-generation analytics tools for basketball.",
        }
    ]
}

​
Building a Graph

Based on this data, we can create a Vector Graph RAG with the following nodes and edges: Nodes:
  • Professor Node with properties name, title, page, bio
  • Research Area Node with properties area and description
  • Department Node with the property name
  • University Node with the property name
  • Lab Node with properties name, research_focus
Vector Nodes for Embeddings: We will have an embedding of the professors’s combined string on their research areas and research description.
  • Professor Combined Research Area and Description Node with the property areas_and_descriptions
Edges:
  • Professor to Research Area Edge
  • Professor to Department Edge
  • Professor to University Edge
  • Professor to Lab Edge
  • Professor to Combined Research Area and Description Embedding

​
Why do we need a Vector Graph RAG?

Let’s imagine it on a larger scale where we have a lot of data on professors, research areas, departments, academic achievements, and labs across 1000+ universities. We can utilize a graph to connect these nodes and edges to each other so that we can traverse the graph faster to find all professors that are working in a specific university, department, research area, or lab e.g we can filter the graph to find all professors that are working in the University X using WHERE in HelixQL. And we can utilize the vector embeddings to find professors that are most similar to a given query e.g “Which professors have worked in a startup before?”.