How Precina Health Uses Memgraph and GraphRAG to Revolutionize Type 2 Diabetes Care with Real-Time Insights
In a recent community call, Josiah Bryan, CTO of Precina Health, walked us through how they use GraphRAG to help patients with Type 2 diabetes. Spoiler alert: they’re crushing it with an approach that scales behavioral care and AI-powered insights for rural and low-income patients.
Watch the full webinar recording – Optimizing Insulin Management: The Role of GraphRAG in Patient Care.
In the meantime, here are the key talking points from the webinar.
Talking Point 1: Why RAG and GraphRAG Matter
Large Language Models (LLMs) like GPT are powerful, but have limits—particularly around context size. Retrieving vast amounts of specific data, especially in something as critical as healthcare, requires better tools. That’s where Retrieval-Augmented Generation (RAG) comes in. Combine it with graph databases, and you get GraphRAG, which allows you to efficiently retrieve and reason through complex, evolving data.
Talking Point 2: Precina Health Win: 1% HbA1C Reduction Every Month
Precina Health isn't just throwing AI at the wall to see what sticks. They’ve systematized Type 2 diabetes management, getting patients to reduce their hemoglobin A1C (HbA1C) by 1% per month (compared to typical annual reductions). Their approach factors in everything—clinical care, social determinants, and behavioral insights.
Talking Point 3: Tech as a Catalyst, Not the Hero
Precina’s approach highlights something missed in tech-driven healthcare: technology is a tool, not the solution. Their system makes small, patient-specific adjustments using AI, but the provider-patient interaction drives real change.
Talking Point 4: How GraphRAG Powers Their System
At its core, Precina’s system, dubbed P3C (Provider-Patient CoPilot), uses Memgraph to manage patient data and vector databases to retrieve insights. It’s not just about medical records; they’re building a knowledge graph that factors in behavioral patterns and social context, allowing providers to offer more personalized care.
Talking Point 5: Memgraph + Vector Search = Efficient Data Retrieval
Precina combines Memgraph’s real-time graph processing with vector search via Qdrant. This lets them perform “multi-hop” reasoning—essentially following complex relationships in patient data to make better care decisions. Vector search finds relevant nodes fast, while Memgraph’s graph structure connects the dots.
Talking Point 6: Real-Time, Personalized Care at Scale
In practical terms, the system helps providers respond to real-time patient inputs. Whether it’s insulin levels or a patient’s emotional state, GraphRAG delivers actionable insights. This is where it gets interesting: it’s not just medical data. Everything from “my bus was late” to “my dog died” gets factored into how providers engage with patients.
Talking Point 7: The Future: GraphRAG, but for Everything
Precina’s current use case is Type 2 diabetes, but Josiah hinted at broader applications for cardiovascular and pulmonary diseases. Their method of combining LLMs with graph databases could easily extend beyond healthcare into other domains where data complexity and relevance are king.
Talking Point 8: Why Memgraph?
When asked why they chose Memgraph, Josiah didn’t hold back: the developer experience. Between strong documentation, support, and flexible deployment (local, Docker, Kubernetes), Memgraph gave them the agility to build and iterate quickly—something Neo4j couldn’t offer.
Q&A
We’ve compiled the questions and answers from the community call Q&A session. Note that we’ve paraphrased them slightly for brevity. For complete details, watch the entire video.
-
What types of use cases work well with GraphRAG, and are there scenarios where it might not be as effective?
- Josiah: GraphRAG is excellent for use cases where you need multi-hop reasoning and relational context, like healthcare or any scenario with complex, evolving data. It's perfect when you need to trace connections between different data points—whether it’s patient records, behavioral insights, or medical treatments. Where it struggles, though, is with long-context generation over massive documents. For example, if you have a 100-page report and want to generate context or pull from multiple documents, GraphRAG can fall short without a sophisticated orchestration system. It’s better suited for retrieving relevant chunks of data rather than handling large in/large out workflows.
-
How exactly does GraphRAG help optimize insulin management, especially since AI isn’t prescribing or adjusting insulin directly?
- Josiah: Our system doesn’t prescribe or tweak insulin doses directly. What GraphRAG does is optimize the overall management of patient care. We feed real-time behavioral and contextual insights to providers, helping them understand a patient’s current situation better. For example, suppose a patient reports having trouble with transportation or mentions a life event like their dog passing away. In that case, that data helps the provider assess how it might impact the patient's compliance with their insulin routine. GraphRAG gives providers a more holistic view of the patient, allowing them to offer more tailored, effective care.
-
Could you achieve the same results using just vector search without GraphRAG? What does GraphRAG add?
- Josiah: Vector search is useful, but it has limitations. With GraphRAG, the advantage comes from handling multiple data hops and relationships. A pure vector search might give you chunks of relevant data, but it won’t capture the connections between those chunks as effectively as a graph database can. In healthcare, for instance, understanding how one patient's condition relates to another or their medical history ties into treatment outcomes requires that relational context. The graph structure helps us connect the dots quickly across different data sets, making the retrieval much richer and more insightful than vector search alone.
-
How do you ensure the explainability of the results when using GraphRAG? Can users understand why certain answers are provided?
- Josiah: I don't typically show users the graph structure or edges directly—that's more for us developers to debug and refine. What I do provide are citations from the source data. So, if the system delivers an answer, it’s backed by references to the original documents or records, which adds transparency. It shows exactly where the information came from and allows users to trace the answer back to specific pieces of data. In practice, these citations are usually enough to satisfy even the most detail-oriented stakeholders.
Further Reading
- Blog post: Building GenAI Applications with Memgraph: Easy Integration with GPT and Llama
- Blog post: How Microchip Uses Memgraph’s Knowledge Graphs to Optimize LLM Chatbots
- Webinar recording: Microchip Optimizes LLM Chatbot with RAG and a Knowledge Graph
- Whitepaper: Enhancing LLM Chatbot Efficiency with GraphRAG (GenAI/LLMs)
Memgraph Academy
If you are new to the GraphRAG scene, we suggest you check out a few short and easy to follow lessons from our subject matter experts. For free. Start with:
