Polyglot Persistence- Not just an exotic idea anymore

Siddhant Agarwal, Developer Relations Lead for APAC at Neo4j, unravels the helices of modern databases- combing through strands like open source environments, LLMs, HTAP, vector data, multi-modal evolution, AI, and a new world ‘with-but-more-than-just relational databases’. Let’s follow this thread.

How has the database landscape been influenced by changes around open-source movements, new LLM formats, and relational databases?

Over the last decade, the database landscape has moved from ‘one-size-fits-all relational’ to a rich ecosystem of specialised data platforms, and three forces are largely driving this shift: open source, AI/GenAI/LLMs, and the limitations of traditional relational thinking.

Open-source movements have lowered the barrier to experimentation. Teams can try graph, time-series, columnar/OLAP, vector, and document databases without a big upfront license decision. These open-source alternatives have not only accelerated innovation but have also made ‘polyglot persistence’ by using the right database for the right job. Not just an exotic idea, but a mainstream practice.

What has changed with new LLMs?

Newer LLMs and the rapidly evolving GenAI industry have completely changed expectations. We now assume that applications should be able to understand text, images, code, logs, documents, and conversations. That expectation pushes data platforms to support vectors, graph structures, and unstructured data natively, not just rows and columns. You see this in the rise of vector indexes, graph + vector hybrids, and Retrieval-Augmented Generation (RAG) architectures, where the database becomes part of the AI pipeline, not just a back-end store.

Are relational databases fading then?

Relational databases aren’t going away, though. They remain the backbone for transactional workloads, but they’re no longer the only default. They’re being complemented by graph databases for connected data and reasoning, by vector databases for semantic search, and by cloud data warehouses and lakehouses for analytical scale. The ‘influence’ is essentially this: we’ve moved from a monoculture to an ecosystem where relational is one important piece, not the whole story.

What new changes can be expected with the rise of motion data, vector formats, multimodal and unstructured data, cross-data querying and HTAP/transyltical databases?

We’re entering a phase where three things converge: real-time signals (motion data), semantic representations (vectors), and rich context (graphs, documents, events). That convergence will drive a few clear shifts. Like Native support for vectors + graph + text in the same engine: The future is not ‘just a vector database’ or ‘just a graph’. It is rather the ability to combine similarity search, graph traversal, and traditional queries within a single logical system. For example, “Find similar products (vector), then follow relationships to suppliers (graph), then filter by regional constraints (relational).”

You can see these new patterns in the rise of vector indexes, graph + vector hybrids, and Retrieval-Augmented Generation (RAG) architectures, where the database becomes part of the AI pipeline, not just a back-end store.

There is also the change of HTAP / translytical patterns by default. The traditional split of OLTP for transactions and OLAP for analytics is being softened. With HTAP/translytical databases, you can run analytical and AI workloads closer to the operational data, with fewer pipelines and less lag. That’s important for LLM-powered applications that require fresh context, not yesterday’s snapshot.

Are multi-modal data shifts serious ones as well?

Multimodal and unstructured data are first-class citizens. Images, audio, PDFs, logs, and events will no longer be ‘dumped in a data lake and forgotten’. Databases are evolving to store the raw assets, as well as derived representations: embeddings, entities, and relationships. That makes multimodal search and reasoning (e.g., “show me all documents, conversations, and tickets related to this customer”) achievable. Also, observe the change in cross-data querying and federation. As organisations accept that a single monolithic database is unrealistic, we’ll see more intelligent cross-source querying: combining vectors from one system, graphs from another, and tabular data from a warehouse. The key change is moving from ‘ETL into one place to ‘orchestrated query across many places’.

In short, the databases that win will be those that make it easy to blend motion data, vectors, graphs, and relational queries into one coherent experience for developers and data teams.

Are databases evolving well to get set for the AI age? Are context, scale and availability still factors to reckon with?

Yes, databases are evolving, and they are evolving pretty rapidly. Modern databases are increasingly incorporating GenAI-native capabilities directly into their core engines. Features like vector indexing, graph reasoning, semantic search, document ingestion, metadata extraction, multimodal embeddings, and AI-assisted querying are no longer ‘add-ons’; they’re becoming first-class architectural requirements. For AI applications, three things still matter the most: context, scale, and reliability and databases are adapting to meet all three.

While databases are evolving well for the AI age, the pressure is high. The systems that successfully combine AI-centric capabilities (vectors, graph reasoning, search) with ‘boring but critical’ capabilities (availability, durability, compliance) will be the ones that enterprises trust.

So, what’s actually the future of relational databases?

Relational databases are not dying; they’re simply being re-scoped. Their future lies in becoming deeply specialised for core transactional workloads such as banking systems, ERP platforms, inventory management, and financial ledgers, where strict consistency and transactional guarantees are essential. Rather than expanding into every new data paradigm, relational systems will increasingly integrate with graph, vector, and analytic engines, playing the role of the system of record while other specialised databases become the system of intelligence.

SQL will continue to serve as the lingua franca for data access across many systems, even as the underlying engines evolve to include graph structures, columnar processing, and vector search. In essence, relational databases will remain foundational, but they will no longer be the sole or primary environment where ‘smart’ behaviour emerges. The intelligence layer will increasingly come from knowledge graphs, AI models, and event streams that operate alongside relational systems, forming a more connected and capable data ecosystem.

Are data virtualisation, federation and anonymisation going to get easier and stronger ahead?

These capabilities have to advance because, without them, most AI strategies will stall on governance and integration challenges. Data virtualisation and federation are becoming more intelligent and increasingly AI-assisted. Instead of manually building every mapping, organisations will rely on tools that use schema inference, entity resolution, and even LLM-driven suggestions to automatically propose joins, mappings, and unified views across disparate systems. This significantly shortens the journey from having dozens of fragmented data sources to being able to query them as a single logical knowledge layer.

Relational databases aren’t going away. They remain the backbone for transactional workloads, but they’re no longer the only default.

At the same time, anonymisation and privacy-preserving techniques are maturing, with stronger support for tokenisation, masking, differential privacy, and the generation of synthetic data. For AI workloads, this allows teams to use realistic data safely for experimentation and model training without exposing sensitive information. Overall, the direction is clear: organisations want a unified, query-friendly, and secure logical data layer. Virtualisation and federation provide the unified view, while anonymisation and access controls ensure it can be used responsibly.

Can you share something about your key customers, their adoption patterns and highlights, and your roadmap ahead?

One of the things we do really well at Neo4j is operate from the customer backwards. We have a strong set of global customers, including British Telecom, DXC, IBM, Klarna, Adobe, Cisco, Citrix and Klarna. Our roadmap in India is to replicate this success by partnering with forward-looking enterprises and building Klarna-like models for at least a few key customers. We are already seeing strong interest in these types of high-impact, unified-experience use cases, and we believe India is poised to become one of the most exciting growth markets for graph innovation.

pratimah@cybermedia.co.in