AI’s institutional memory gap sparks $15M bet on decoupled knowledge layers

The latest wave of enterprise AI investments is targeting a persistent bottleneck: how to keep models aligned with organizations’ ever-changing internal realities.
Instead of relying on pre-trained models that must be repeatedly fine-tuned or rebuilt as business contexts shift, it constructs a dedicated knowledge layer from each customer’s internal data, processes, and terminology. This layer then informs AI responses, allowing organizations to adapt without overhauling their entire infrastructure.
The need for such solutions has become clearer as enterprise AI adoption accelerates. Spending on these systems is projected to reach $177 billion globally this year, yet surveys indicate most organizations still lack the data infrastructure to support large-scale deployments effectively. A recent study of 10,000 businesses across 32 countries found that fewer than one in ten considered their data fully prepared for AI, while less than a third reported strong returns from their investments. For industries where precision and compliance are critical—such as financial services, cybersecurity, or healthcare—this misalignment carries significant operational and regulatory risks.
At the core of the platform is a GPU-accelerated knowledge graph designed to dynamically map relationships between structured and unstructured enterprise data. Unlike traditional systems that require predefined schemas, this graph builds connections automatically as new information is ingested. The company claims it can adjust up to 10% of an AI model’s weights through targeted updates, keeping responses accurate as business rules, internal terminology, or compliance requirements evolve—without necessitating full retraining cycles.
This architecture also introduces model portability. If an organization switches AI providers, its knowledge graph can theoretically support compatible alternatives without requiring a complete rebuild. That’s a notable departure from current enterprise deployments, where model changes often trigger lengthy and costly engineering efforts.
Flexibility in deployment is another key advantage. The platform supports multi-cloud environments (AWS, Azure, Google Cloud) and on-premises or air-gapped configurations, catering to sectors with strict data residency requirements. The company has already earned AWS’s AI Software Competency designation and was selected for Amazon’s Global Startup Program earlier this year.
The funding will accelerate development of automated data pipelines that unify internal documents, business records, and operational datasets into a single knowledge layer.
While other startups have pursued related challenges—such as Ema’s workflow automation or RapidCanvas’s agent-based systems—the focus on treating institutional knowledge as a standalone, updatable asset reflects a broader industry reckoning. As cloud providers embed AI capabilities deeper into their stacks, the ability to maintain control over data while enabling adaptability will likely determine which solutions gain traction in regulated environments.
The question remains whether this approach can scale before enterprises face a trade-off between agility and precision. Watch for customer announcements in high-stakes sectors—particularly those with stringent compliance needs—as indicators of adoption.
Sources: pulse2.com
“The funding suggests a shift toward architectures that treat enterprise knowledge as a separate, updatable asset—rather than a model’s embedded feature—amid growing frustration with deployment drift.”
Read the original reporting
The outlets below did the original reporting.
Related briefs
- Nvidia invests in Simplismart’s $14M round as enterprises seek AI cost controls
- Mintive raises $7M to secure AI agents with identity and permissions
- Infisical raises $40M at $600M valuation to secure AI agent access
- Nvidia launches AI factory blueprints, agent safety tools at GTC Berlin
- Nvidia reframes AI as a factory and agents as the workload
This brief was drafted automatically from the sources above and published under our editorial policy. Spotted an error? Tell us.