At 1Optic we think about enterprise AI as a pyramid and like any pyramid, what holds it up matters far more than what sits on top. We have written before about why we don’t believe in AI heaven. This post explains the architecture behind that conviction: two distinct halves, one shared structure, and a very deliberate boundary between them.
Foundation first: the single pane of glass
The base of the pyramid is the 1Optic data platform: a robust, unified database that combines data from across the entire organization into one coherent, queryable layer. CRM records, operational logs, inventory data, sensor feeds, legacy and multi-age metrics: everything flows in, everything is reconciled, and every downstream consumer, human analyst or AI model, works from the same version of truth. This is harder than it sounds. Most organizations have data sprawl baked deep into their architecture.
Data teams spend the majority of their time not doing analysis, but finding, cleaning, and reconciling data that was never designed to coexist. The single pane of glass is not just a convenience it is the non-negotiable prerequisite for everything above it. Once the foundation is solid, the second layer focuses on preparing data for AI consumption. Raw unified data and AI-ready data are not the same thing. Between them lies structuring for machine consumption, labeling pipelines, context enrichment, and access controls that let models learn from sensitive data without exposing it. This is where 1Optic’s data specialists work: in close collaboration with the customer’s domain experts. Because no one understands your data better than the people who created it, and no one understands how to prepare it for AI better than specialists who have done it across many domains and systems.
Embeddings: where data becomes meaning
At the boundary between the data part and the AI part of the pyramid sits the most important layer: embeddings. An embedding is a dense numerical vector that captures the meaning and relationships within a piece of data. Text, structured records, time-series, even user behavior: all of it can be transformed into embeddings that encode semantic similarity in mathematical space. 1Optic generates embeddings directly from the prepared data beneath. This is the critical distinction. Generic pre-trained models produce generic embeddings they know language in general, but not your language.
Embeddings generated from your own data know your product catalog, your customer vocabulary, your operational patterns. They are specific in a way that generic models simply cannot replicate. These embeddings sit at the boundary because they bridge both worlds. They are the output of the data work below clean, prepared, enriched and the input to the AI work above. They can be fed directly into AI modules for further processing: retrieval-augmented generation, semantic search, anomaly detection, recommendation, and more.
The apex: modular, composable, yours
The AI part of the pyramid sits above the embedding layer. Because the embedding foundation is standardised and high quality, the AI modules at the apex can be assembled modularly, mixing and matching models, vendors, and approaches without rebuilding the plumbing each time. Multi-vendor environments stop being a liability and become a feature. Different teams can work at the apex with confidence, knowing the data beneath them is consistent and trustworthy. This is what 1Optic prepares data for: AI teams working at the apex of the pyramid, with the full power of a unified, embedded, semantically rich data layer underneath them.
Sustainable progress, step by step
As we have said in earlier posts: we do not promise AI heaven. We believe in sustainable progress improving AI step by step, in a continuous loop of collaboration between 1Optic’s data specialists and the customer’s domain experts. The pyramid is that loop made visible. Each layer builds on the one below it. Each improvement to the foundation amplifies everything above it. There is no shortcut from raw data to AI value but there is a clear, proven path. Build the base. Prepare the layers. Generate meaning from what you know. Then improve it, together, over time.