🗺️System of Record

Shriram Sridharan, Co-founder/Engineering Lead

From Cloud native to AI native

I began my career at AWS, where I focused on building cloud-native relational database systems. The cloud era introduced unprecedented elasticity, abundance, and scale, revolutionizing how we design and deploy systems. This presented an opportunity to reimagine how enterprise data systems have been built with this changed paradigm. I was fortunate to be part of a stellar team that reimagined how relational databases could be built for the cloud. Relational databases were primarily built with scarcity of the disk or the compute whereas the cloud changed that fundamental assumption. Disk/Compute was available in abundance where as the scarce commodity was the network. Amazon Aurora was built with this paradigm shift to be a cloud-native relational database system[1][2]. I saw the service scale pre-launch with $0 in revenue to becoming the fastest growing service in all of AWS during my stint there. At my next stint in Confluent, we did the same and re-imagined Kafka as a cloud-native event-streaming system [3][4]

Today, we stand at the forefront of another paradigm shift—AI. With AI, the computing model has evolved from containers to models, and memory management has transitioned from virtual memory to context windows. Large Language Models (LLMs) have the ability to understand both structured and unstructured data alike with a natural language interface and have the ability to reason by incorporating external tools. This shift presents another opportunity to reimagine enterprise data systems, particularly ones that have been built with a rigid structure in mind. There is the ability now to unify data from multiple different data sources and extract insights and valuable information which was not possible previously. At Rox, we are building an AI-native enterprise system with this paradigm shift as first class principles.

The Change

During my time at Confluent, I witnessed the transformative shift from a commit-to-consumption business model and how data evolved from residing in enterprise systems to being unified in data warehouses. This transition allowed organizations to construct a single representation of their data, powering BI dashboards for revenue teams. Scaled B2B SaaS companies mirrored this approach, building in-house revenue stacks atop structured data to drive business outcomes.

However, through conversations with Account Executives (AE), we realized a significant gap in how workflows were managed. Despite unified structured data systems, AEs relied on a plethora of tools for tasks like research (public internet), contact search (often LinkedIn and other tools), team and org chart insights, meeting records (meeting AI recorders), and more. There was no unified platform—a "single pane of glass"—to seamlessly integrate these disparate tools and deliver insights and perform actions into one place. Additionally, access to data from systems of record has traditionally been limited to dashboards or applications. With AI agents, the delivery of this information must meet users where they are—whether through mobile apps, web interfaces, desktop tools, or even real-time notifications during calls.

This is the vision Rox is set out to build.

Where are we now ?

Over the last few months, we have built a unified System of Record (SoR) layer that provides event-driven access to a knowledge graph constructed from structured, unstructured, and semi-structured data. The SoR is enterprise-ready, supporting secure data sharing while maintaining data in a warehouse-native manner. In secure data-sharing mode, Rox ensures that no raw customer data is stored at rest. External and internal data flows into the Rox Data Lake-house. The Rox Data Lake-house processes multi-modal data at TB+ scale, integrating unstructured sources like news, job postings, and blogs with structured data from CRMs, support systems, and product usage time-series data. For enterprise app data, Rox offers bi-directional API-based connectors for tools like email, calendars, and Slack.

On top of the Data lake-house, we construct a unified knowledge graph which maps to entities such as companies and people across distributed data sources. We built an in-house Knowledge graph construction and storage system to cater to our growing data volumes. Data access to the knowledge graph is via real-time event streams to the agent swarms. AI agents can read from the knowledge graph and write back into the system of record in a seamless manner enabling them to do work on behalf of the AE. We also provided developer-friendly API access externally for enterprises on these APIs [5].

What's next ?

We have a very exciting roadmap ahead of us. Our immediate focus is on enhancing real-time capabilities by optimizing the streaming graph build and reducing API access latency. We aim to expand the connector hub to integrate with more enterprise business systems and app systems to have more data flow in to our data lake to be unified. This also includes sales enablement materials which is private to enterprise orgs. As we onboard more enterprises, the knowledge graph and Rox entity representation will evolve to better reflect complex business relationships. Unlocking hidden data within business systems remains a priority, addressing the synthesis challenge to deliver precise, actionable insights exactly when they’re needed.

From a public data standpoint, we have to broaden our content ingestion pipelines, extending beyond news, job postings, and blogs to include additional sources while improving data quality through advanced entity extraction and unification. Tackling poor data quality with intelligent data wrangling and imputation using LLMs is another key challenge we’re addressing with AI agents. As mentioned before, context window management is critical, requiring thoughtful selection of high-quality data for optimal results in the evolving computing model. Lastly, we are developing hybrid deployment options, including in-VPC solutions, to meet enterprise security and scalability requirements.

At the limit, we aim to get the right to run the world's revenue data in Rox.

This journey wouldn’t be possible without our incredible team—a powerhouse of talent combining years of experience from AWS, StreamSets, Confluent, and Ramp, alongside accomplished researchers, passionate graduates from top institutions, and IOI/IMO gold medalists. Spread across SFO and BLR, they’ve achieved remarkable milestones in just 279 days. I’m beyond proud of what we’ve built together and thrilled for the exciting chapters that lie ahead!

If these problem statements sound interesting to you, feel free to reach out to [e-mail]. You can read more in-depth on the System of Record HERE