RAG pipelines are everywhere, but most fall short when the questions get complex. In this session, I’ll walk through a real-world case study where we hit the limits of pure retrieval while building an AI knowledge base for a large real estate firm.
To solve it, we introduced a Reasoning Layer using Neo4j, an open-source graph database. This helped us model and traverse relationships between projects, vendors, delays, and consultants, unlocking answers that embeddings alone could never find.
We’ll cover
Where RAG breaks down (with examples)
How graphs and Cypher helped us reason over data
The hybrid architecture we used (Vertex AI + Neo4j + LangChain)
Open-source alternatives to commercial solutions
Lessons for building better, explainable AI systems with FOSS tools
If you're working with LLMs and want to move beyond just retrieval, this talk shows how open-source graph tools can add real intelligence to your system. It’s a practical, applied deep dive into reasoning at scale.
Here are some key takeaways
RAG isn’t enough for complex questions.
Retrieval helps fetch context, but it struggles with reasoning, chaining logic, or understanding relationships.
Graphs bring structure and explainability to AI.
Using Neo4j and Cypher, we modeled entities (projects, vendors, delays) and built logic-based insights.
Open-source tools can power enterprise-grade reasoning.
While we used Neo4j, similar results are achievable with Apache AGE, ArangoDB, or RedisGraph.
Hybrid systems outperform standalone models.
Combining retrieval, reasoning, and LLMs gives you smarter, grounded, and more trusted AI outputs.
FOSS graphs aren't just for analytics, they’re the missing layer in AI.
The Reasoning Layer is a force multiplier that’s still underutilized in the open-source GenAI ecosystem.