Blog
Planck AI Dictionary Concept
Planck AI Dictionary Concept
The Context Dictionary
A Semantic Runtime for Deterministic Enterprise AI
Most enterprise AI systems today are built around large language models and retrieval pipelines. A typical architecture includes embeddings, vector search, and prompt templates layered on top of enterprise data sources.
While this architecture works well for document retrieval and summarization, it often fails when applied to operational enterprise data.
The reason is simple: LLMs understand language, but they do not understand business semantics.
They cannot reliably infer how enterprise systems represent entities, how metrics should be calculated, or how datasets relate to each other. As a result, many AI deployments struggle with inconsistent answers, incorrect aggregations, and fragile prompt engineering.
At Planck AI we address this problem by introducing a new architectural layer between data and AI models:
The Context Dictionary.
Why Enterprise AI Needs a Semantic Runtime
Enterprise data systems are not designed for natural language interaction.
Operational data lives across systems such as:
ERP databases
CRM platforms
data warehouses
operational logs
document repositories
Even when these systems expose structured schemas, the meaning of fields is rarely obvious to an AI model.
Consider a simple example.
A database column named price might represent:
unit price
invoice price
discounted price
contract price
Without explicit semantic definitions, an AI system must infer the correct interpretation from context. In practice, this often leads to incorrect answers.
Traditional BI systems solved this problem through semantic layers that define metrics and dimensions. However, those systems were not designed to guide AI reasoning or query interpretation.
Modern AI systems therefore require something more powerful:
A machine-readable representation of business semantics that can guide retrieval, reasoning, and execution.
The Context Dictionary Architecture
The Context Dictionary is a structured artifact that encodes the semantic structure of enterprise data.
It acts as a semantic runtime environment for AI systems.
Conceptually, it combines several layers of functionality that historically existed in separate systems:
Capability | Traditional System |
|---|---|
Metric definitions | Looker / dbt semantic layers |
Contextual retrieval | Vector databases like Pinecone |
Entity relationships | Knowledge graphs |
Query guidance | Prompt engineering |
The Context Dictionary unifies these capabilities into a single structured specification that AI systems can interpret programmatically.
Core Components
A typical Context Dictionary contains several categories of metadata.
Entities
Entities represent the core objects within a system.
Examples include:
products
customers
invoices
fiber network segments
Each entity defines its primary key and its relationships with other entities.
Example:
entity: productsprimary_key: skurelationships: - orders.product_id - inventory.sku
entity: productsprimary_key: skurelationships: - orders.product_id - inventory.sku
entity: productsprimary_key: skurelationships: - orders.product_id - inventory.sku
This structure enables AI systems to understand how datasets relate to each other.
Dimensions
Dimensions represent descriptive attributes used for grouping and filtering.
Examples include:
product category
region
supplier
store location
Dimensions include additional semantic hints such as:
cardinality
hierarchical structure
ordering rules
Example:
dimension: categorysemantic_type: dimensioncardinality_hint: mediumhierarchy: - category_level_1 - category_level_2 - leaf_category
dimension: categorysemantic_type: dimensioncardinality_hint: mediumhierarchy: - category_level_1 - category_level_2 - leaf_category
dimension: categorysemantic_type: dimensioncardinality_hint: mediumhierarchy: - category_level_1 - category_level_2 - leaf_category
These hints guide query planning and visualization.
Metrics
Metrics represent quantitative measurements.
Examples include:
revenue
inventory levels
delivery time
network fault rate
Each metric includes aggregation rules and formatting semantics.
Example:
metric: pricesemantic_type: metricunit: currencypreferred_aggregation: sumrecommended_aggregations: - sum - avg - min - max
metric: pricesemantic_type: metricunit: currencypreferred_aggregation: sumrecommended_aggregations: - sum - avg - min - max
metric: pricesemantic_type: metricunit: currencypreferred_aggregation: sumrecommended_aggregations: - sum - avg - min - max
This ensures that queries involving metrics are interpreted consistently.
Parsing Rules
Many enterprise datasets encode hierarchical or composite structures in single fields.
For example, product categories might be stored as path strings such as:
Safety Supplies/Apparel/Vests
Safety Supplies/Apparel/Vests
Safety Supplies/Apparel/Vests
The Context Dictionary defines how these fields should be parsed.
Example:
parsing: delimiter: "/" produce_fields: - category_level_1 - category_level_2 - leaf_category
parsing: delimiter: "/" produce_fields: - category_level_1 - category_level_2 - leaf_category
parsing: delimiter: "/" produce_fields: - category_level_1 - category_level_2 - leaf_category
This allows AI systems to reason over hierarchical structures without manual transformation.
Visualization Semantics
Another unique aspect of the Context Dictionary is that it encodes visualization semantics.
This allows AI systems to generate structured visual outputs rather than raw text.
Example:
chart_semantics: preferred_chart_types: - line - bar - histogram
chart_semantics: preferred_chart_types: - line - bar - histogram
chart_semantics: preferred_chart_types: - line - bar - histogram
For example:
time series fields default to line charts
categorical breakdowns default to bar charts
boolean states may use donut charts
This allows AI responses to include dashboards and analytics visualizations automatically.
Query Resolution Workflow
When a user submits a natural language query, the Planck AI system resolves it using the Context Dictionary.
The workflow typically follows these steps:
Query Interpretation
The system maps natural language terms to dictionary entities, metrics, and dimensions.
Example:
"top products by revenue""top products by revenue""top products by revenue"is interpreted as:
entity: productsmetric: revenueaggregation: sumgroup_by: product_name
entity: productsmetric: revenueaggregation: sumgroup_by: product_name
entity: productsmetric: revenueaggregation: sumgroup_by: product_name
Query Planning
The dictionary provides guidance for:
which dataset to query
how to join tables
which aggregations to apply
Execution
The query is executed through the Verified Compute Layer, ensuring deterministic results.
Response Generation
The system returns:
structured results
appropriate visualizations
contextual explanations
Because the interpretation is grounded in dictionary semantics, identical queries produce consistent results.
Deterministic AI
One of the primary benefits of the Context Dictionary is determinism.
Traditional LLM systems operate probabilistically. Two identical prompts may produce slightly different interpretations.
By grounding AI reasoning in structured semantic definitions, the Context Dictionary reduces ambiguity.
This allows enterprise AI systems to provide answers that are:
reproducible
auditable
consistent across queries
For operational systems and regulated industries, this reliability is essential.
Beyond Retrieval-Augmented Generation
Retrieval-Augmented Generation (RAG) systems focus primarily on retrieving relevant text.
However, many enterprise use cases require reasoning over structured operational data.
Examples include:
supply chain monitoring
financial KPI analysis
telecom network performance diagnostics
ERP operational reporting
These problems require more than document retrieval.
They require structured reasoning over business data models.
The Context Dictionary enables this by providing AI systems with a machine-readable understanding of enterprise data semantics.
The Context Dictionary as AI Infrastructure
In the broader Planck AI architecture, the Context Dictionary functions as the foundational semantic layer.
It powers several higher-level components:
the Resolution Engine, which interprets user queries
the Verified Compute Layer, which executes deterministic calculations
the Execution Layer, which triggers workflows and operational actions
By separating semantic understanding from model inference, Planck AI enables enterprises to build reliable AI applications on top of their existing data infrastructure.
Toward a Semantic Runtime for AI
As enterprise AI systems evolve, it is becoming increasingly clear that models alone are not enough.
Reliable AI requires structured representations of business knowledge.
The Context Dictionary provides that structure.
It transforms enterprise data into a semantic runtime that AI systems can interpret consistently.
In doing so, it enables a new generation of enterprise applications where natural language becomes a reliable interface to complex operational systems.
Planck AI is building the infrastructure that makes this possible.
The Context Dictionary
A Semantic Runtime for Deterministic Enterprise AI
Most enterprise AI systems today are built around large language models and retrieval pipelines. A typical architecture includes embeddings, vector search, and prompt templates layered on top of enterprise data sources.
While this architecture works well for document retrieval and summarization, it often fails when applied to operational enterprise data.
The reason is simple: LLMs understand language, but they do not understand business semantics.
They cannot reliably infer how enterprise systems represent entities, how metrics should be calculated, or how datasets relate to each other. As a result, many AI deployments struggle with inconsistent answers, incorrect aggregations, and fragile prompt engineering.
At Planck AI we address this problem by introducing a new architectural layer between data and AI models:
The Context Dictionary.
Why Enterprise AI Needs a Semantic Runtime
Enterprise data systems are not designed for natural language interaction.
Operational data lives across systems such as:
ERP databases
CRM platforms
data warehouses
operational logs
document repositories
Even when these systems expose structured schemas, the meaning of fields is rarely obvious to an AI model.
Consider a simple example.
A database column named price might represent:
unit price
invoice price
discounted price
contract price
Without explicit semantic definitions, an AI system must infer the correct interpretation from context. In practice, this often leads to incorrect answers.
Traditional BI systems solved this problem through semantic layers that define metrics and dimensions. However, those systems were not designed to guide AI reasoning or query interpretation.
Modern AI systems therefore require something more powerful:
A machine-readable representation of business semantics that can guide retrieval, reasoning, and execution.
The Context Dictionary Architecture
The Context Dictionary is a structured artifact that encodes the semantic structure of enterprise data.
It acts as a semantic runtime environment for AI systems.
Conceptually, it combines several layers of functionality that historically existed in separate systems:
Capability | Traditional System |
|---|---|
Metric definitions | Looker / dbt semantic layers |
Contextual retrieval | Vector databases like Pinecone |
Entity relationships | Knowledge graphs |
Query guidance | Prompt engineering |
The Context Dictionary unifies these capabilities into a single structured specification that AI systems can interpret programmatically.
Core Components
A typical Context Dictionary contains several categories of metadata.
Entities
Entities represent the core objects within a system.
Examples include:
products
customers
invoices
fiber network segments
Each entity defines its primary key and its relationships with other entities.
Example:
entity: productsprimary_key: skurelationships: - orders.product_id - inventory.sku
This structure enables AI systems to understand how datasets relate to each other.
Dimensions
Dimensions represent descriptive attributes used for grouping and filtering.
Examples include:
product category
region
supplier
store location
Dimensions include additional semantic hints such as:
cardinality
hierarchical structure
ordering rules
Example:
dimension: categorysemantic_type: dimensioncardinality_hint: mediumhierarchy: - category_level_1 - category_level_2 - leaf_category
These hints guide query planning and visualization.
Metrics
Metrics represent quantitative measurements.
Examples include:
revenue
inventory levels
delivery time
network fault rate
Each metric includes aggregation rules and formatting semantics.
Example:
metric: pricesemantic_type: metricunit: currencypreferred_aggregation: sumrecommended_aggregations: - sum - avg - min - max
This ensures that queries involving metrics are interpreted consistently.
Parsing Rules
Many enterprise datasets encode hierarchical or composite structures in single fields.
For example, product categories might be stored as path strings such as:
Safety Supplies/Apparel/Vests
The Context Dictionary defines how these fields should be parsed.
Example:
parsing: delimiter: "/" produce_fields: - category_level_1 - category_level_2 - leaf_category
This allows AI systems to reason over hierarchical structures without manual transformation.
Visualization Semantics
Another unique aspect of the Context Dictionary is that it encodes visualization semantics.
This allows AI systems to generate structured visual outputs rather than raw text.
Example:
chart_semantics: preferred_chart_types: - line - bar - histogram
For example:
time series fields default to line charts
categorical breakdowns default to bar charts
boolean states may use donut charts
This allows AI responses to include dashboards and analytics visualizations automatically.
Query Resolution Workflow
When a user submits a natural language query, the Planck AI system resolves it using the Context Dictionary.
The workflow typically follows these steps:
Query Interpretation
The system maps natural language terms to dictionary entities, metrics, and dimensions.
Example:
"top products by revenue"is interpreted as:
entity: productsmetric: revenueaggregation: sumgroup_by: product_name
Query Planning
The dictionary provides guidance for:
which dataset to query
how to join tables
which aggregations to apply
Execution
The query is executed through the Verified Compute Layer, ensuring deterministic results.
Response Generation
The system returns:
structured results
appropriate visualizations
contextual explanations
Because the interpretation is grounded in dictionary semantics, identical queries produce consistent results.
Deterministic AI
One of the primary benefits of the Context Dictionary is determinism.
Traditional LLM systems operate probabilistically. Two identical prompts may produce slightly different interpretations.
By grounding AI reasoning in structured semantic definitions, the Context Dictionary reduces ambiguity.
This allows enterprise AI systems to provide answers that are:
reproducible
auditable
consistent across queries
For operational systems and regulated industries, this reliability is essential.
Beyond Retrieval-Augmented Generation
Retrieval-Augmented Generation (RAG) systems focus primarily on retrieving relevant text.
However, many enterprise use cases require reasoning over structured operational data.
Examples include:
supply chain monitoring
financial KPI analysis
telecom network performance diagnostics
ERP operational reporting
These problems require more than document retrieval.
They require structured reasoning over business data models.
The Context Dictionary enables this by providing AI systems with a machine-readable understanding of enterprise data semantics.
The Context Dictionary as AI Infrastructure
In the broader Planck AI architecture, the Context Dictionary functions as the foundational semantic layer.
It powers several higher-level components:
the Resolution Engine, which interprets user queries
the Verified Compute Layer, which executes deterministic calculations
the Execution Layer, which triggers workflows and operational actions
By separating semantic understanding from model inference, Planck AI enables enterprises to build reliable AI applications on top of their existing data infrastructure.
Toward a Semantic Runtime for AI
As enterprise AI systems evolve, it is becoming increasingly clear that models alone are not enough.
Reliable AI requires structured representations of business knowledge.
The Context Dictionary provides that structure.
It transforms enterprise data into a semantic runtime that AI systems can interpret consistently.
In doing so, it enables a new generation of enterprise applications where natural language becomes a reliable interface to complex operational systems.
Planck AI is building the infrastructure that makes this possible.
Back to Blogs