Welcome to Tursio

Tursio is an AI-native query engine for search and analysis that helps you work with structured data faster, more simply, and more accurately. This guide is designed to help you get the most out of Tursio by describing its key features, walking through step-by-step workflows, and illustrating sample scenarios that mirror real-world business needs. Whether you are searching across complex datasets, applying advanced filters, or generating insights on the fly, this documentation will give you the clarity and examples you need to become productive with Tursio quickly.

Why Tursio?

Generative AI is unleashing a new generation of go-getters who believe in getting things done faster than ever before. Most success stories leverage pre-trained large language models (LLMs) for personal productivity tasks such as writing, designing, coding, reporting, and web search. However, when it comes to business applications — analyzing products, supporting customers, or running operations — pre-trained models fall short. These tasks depend on structured enterprise data, typically stored as systems of record in relational databases, which generic AI models cannot access or understand easily. Retrieving relevant pieces of information from structured databases, i.e., enterprise search, is critical to supporting these tasks.

What about traditional BI? Traditional business intelligence (BI) has long been the standard for analyzing structured data, with tools like Power BI, ThoughtSpot, and Tableau leading the way. This approach relies on data experts to first translate business questions into complex data queries and then build dashboards and charts to visualize the results. Business users, in turn, interact with these dashboards to find the insights they need. Over the past two decades, this process has become the norm in most organizations, and it rests on two key assumptions: first, that business users do not have direct access to data, and second, that waiting hours or even days for answers is acceptable. Modern organizations are challenging both assumptions.

Towards "Perplexity" for Structured Data. Perplexity has shown that people prefer clear, concise answers over a list of web links. In the same way, today's business users are not looking for dashboards or charts — they want straightforward answers to their business questions, delivered quickly and effortlessly.

Tursio bridges this gap by offering a search interface that delivers direct answers from structured data. It begins by accurately retrieving relevant data through auto-prompting, which guides business users to understand what data is available. It then applies reasoning to generate precise answers based on the retrieved data. Beyond just answers, Tursio also provides supporting artifacts for deeper exploration and comprehensive analysis.

Much like Perplexity does for the web, Tursio emphasizes identifying the right data sources and their necessary transformations before addressing the question. With Tursio, businesses no longer need to worry about data correctness — every answer is grounded in a robust, transparent query plan, delivering consistent and reliable results every time.

Key Differentiators

• Guaranteed Data Accuracy: Tursio ensures 100% accurate data retrieval, grounding AI-generated answers in precise, reliable information.

• Business User Friendly: Designed for users who may not know how to formulate data queries but can articulate their business questions once the relevant data is surfaced.

• Automated Semantic Modeling: Automatically builds a well-defined semantic model of your data, which guides the generation of accurate and efficient query plans.

• Consistent, Repeatable Results: Delivers the exact same data for the same question every time, ensuring consistency and trust in answers.

• Persistent Data Context: Maintains full data context throughout the answer generation process, unlike chatbots that may lose context during extended interactions.

• Unified Search Experience: Offers a single search box to connect and query across all structured databases, streamlining access to enterprise data.

• Separation of Data and AI: Disaggregates data storage and AI processing, eliminating vendor lock-in and enabling independent scaling of data and AI infrastructure.

Value Proposition

Accuracy is critical with enterprise data. Incorrect answers can have severe consequences, yet it is hard for most business users to spot data errors or fix SQL queries. Enterprise data in databases and data warehouses is complex, and most users are non-experts — they are not aware of the schemas, semantics, and other metadata — making it incredibly hard for them to query it accurately.

State-of-the-art AI for text-to-SQL has accuracy hovering under 80% on the BIRD-SQL benchmark. Interestingly, improvements seem to have plateaued over the last year and a half.

Things are unlikely to change substantially with newer LLMs, since pre-training is already hitting the wall. The current approach to the accuracy problem relies on extensive manual prompting, hoping to provide more relevant context for more accurate results. This creates an adoption challenge when moving from the traditional BI world — with well-curated semantic models and consistent results — to the new AI world of probabilistic, non-deterministic outputs. There is also a newer trend of building the semantic model in the data platform itself, e.g., semantic views in Snowflake and knowledge store in Databricks. However, this does not simplify the effort or expertise needed to curate the model.

Tursio removes the burden of prompt construction from users and instead gives them a vibe coding experience with auto-completions. The idea is to infer a lightweight semantic model automatically and guide users to the right questions step by step, while ensuring correct operator trees behind the scenes. The figure below illustrates the key idea in the Tursio AI engine.

Users can access the space of all possible entities, but the moment they specify "show total extended price", it creates an aggregate operator on the Lineitem table (Step 1). Now the users can see all other valid entities that Lineitem can link to, based on the semantic model. Once they specify "by order status" as a grouping field, the operator tree automatically adds the corresponding join operation with the Order table. Again, the user sees the list of reachable entities, and when they apply a filter on "order priority", it is pushed into the operator tree. Users can continue building arbitrarily complex queries in natural language, while Tursio takes care of constructing the valid operator tree behind the scenes. Once they are done, Tursio converts the operator tree into the backend SQL dialect for execution.

Solution Architecture

Tursio presents an enterprise-ready architecture with data security, semantic understanding, and multiple integration options. The core architecture principles are:

1. No Data Movement. Tursio operates on a zero-data-movement principle. The platform does not copy, transfer, or store customer data in external locations. Instead, Tursio connects directly to existing data sources and performs analysis in place, ensuring data sovereignty and compliance with strict security requirements. This approach eliminates data residency concerns and reduces the security risks associated with data replication.

2. Semantic Model Building. Tursio builds domain-specific semantic models that understand the business context and terminology of each organization. The platform:

   • Interprets business semantics through domain ontology awareness
   • Creates constant feedback loops to improve model accuracy
   • Develops targeted AI models specialized for specific data domains
   • Maintains context awareness across different business functions
3. LLM-Powered Operations. Tursio leverages multiple large language model (LLM) calls for various operations:
   • Infer Relationships: Identify and map connections between entities within structured data
   • Build Semantic Model: Construct semantic models by interpreting business context, mapping terminology, and establishing relationships within structured data
   • Operator Context: Interpret the context of query operations to guide valid interpretation of user queries
   • Query Context: Analyze and enrich the context of user queries to ensure accurate and relevant results
   • Result Post-processing: Summarize, visualize, refine, and format results after initial processing to ensure actionable outputs
4. Security and Audit Framework. Tursio implements comprehensive security and audit controls:
   • Zero Data Movement: Eliminates data exposure risks
   • Access Controls: Role-based permissions and authentication
   • Audit Logging: Complete activity tracking and compliance reporting
   • Compliance: Built-in support for industry-specific regulations
   • Privacy Protection: No data retention beyond analysis sessions
5. Multiple User Interface Options. Tursio provides flexible integration options through various user interfaces:
   • Web Application: Full-featured browser-based interface
   • API Integration: RESTful APIs for custom application integration
   • Embedded Widgets: Components that can be embedded in existing systems
   • Mobile Interfaces: Responsive design for mobile access
   • Desktop Applications: Native applications for specific use cases
   • Chat Interfaces: Conversational AI interfaces for natural interaction

Content Layout

The rest of this documentation is organized as follows:

• Onboarding: Describes how to add database connections, define the scope of querying, and run the training.
• Semantic Modeling: Gives an overview of the semantic model inferred by Tursio and how to extend it.
• Querying: Covers query modes, prompting techniques, result analysis, and sharing capabilities.
• Management: Explains user management, access control mechanisms, and telemetry visibility.
• About Us: Learn more about the company and check out relevant resources.