1Z0-1159-26 · Domain 1 of 6

1Z0-1159-26 Domain 1: Explain Oracle AI Vector Search Foundations

This section covers the building blocks of Oracle AI Vector Search: embeddings, the native VECTOR data type and its formats, generating embeddings with in-database ONNX models or external providers, and the SQL functions that create and inspect vectors.

Exam weight estimate
Not published
Official topics estimate
4
Estimated questions estimate
~5
Suggested study time estimate
6–8 hrs

About 12 min read Reviewed October 2026

Overview

Domain 1 sets up the vocabulary for everything else on 1Z0-1159-26. It covers what a vector embedding is, why semantic similarity search differs from keyword search, and how Oracle AI Database 26ai stores embeddings natively in the VECTOR data type alongside relational, JSON and spatial data. You need to know how embeddings are produced: inside the database with an ONNX model imported through DBMS_VECTOR.LOAD_ONNX_MODEL and called with VECTOR_EMBEDDING, or outside it through a provider such as OCI Generative AI. The section also covers the vector helper functions, including TO_VECTOR, VECTOR_DIMENSION_COUNT and VECTOR_SERIALIZE.

A note on exam status: this page targets the Agentic AI for Oracle AI Database Professional exam (1Z0-1159-26). The site listing describes it as the successor to Oracle AI Vector Search Professional (1Z0-184), but we could not confirm that replacement on an Oracle page, so check the Oracle University catalog before you register.

How it's tested

Foundation questions on this exam usually describe a small design choice and ask which data type, function or embedding approach fits. Expect short SQL fragments, a VECTOR column definition to judge, or a scenario about where embeddings should be generated given security and latency limits.

  • Pick the declaration

    A table needs to store embeddings from a named model, and you choose the VECTOR definition with the right dimension count and number format.

  • Inside or outside the database

    A requirement about data leaving the network or about latency points to an in-database ONNX model or to an external embedding provider.

  • Name the helper function

    You identify which SQL function converts text to a vector, counts dimensions or serializes a vector back to a string.

Reasoning it through

A bank wants embeddings created without sending customer text to any outside service. That rules out calling a hosted provider through UTL_TO_EMBEDDING with a credential. Loading an ONNX embedding model into the database with DBMS_VECTOR.LOAD_ONNX_MODEL and calling VECTOR_EMBEDDING in SQL keeps the text and the model inside the database boundary, so that is the design the question expects.

Topics 4

Official topics from the Oracle outline, in outline order.

1.1 Explain vector embeddings and semantic similarity searchWhat embeddings represent and why similarity replaces exact matching.
What it means
An embedding model turns unstructured content such as text or images into a fixed-length array of numbers. Items with similar meaning end up close together in that vector space, so a search can return conceptually related rows even when no keywords match.
What to understand
Similarity search ranks results by distance rather than filtering on equality. The quality of results depends on the embedding model, not on the database. Every vector you compare must come from the same model, because two models produce incompatible spaces even when the dimension counts happen to match.
How it shows up
Questions contrast semantic search with LIKE or Oracle Text keyword search and ask which one finds paraphrased content.
Common confusion
Thinking the database understands meaning by itself; the embedding model creates the meaning, the database stores and compares it.
1.2 Store embeddings with the VECTOR data typeVECTOR columns, dimensions, formats and sparse vectors.
What it means
Oracle AI Database 26ai adds a native VECTOR data type. A column can be declared with a dimension count and a format such as INT8, FLOAT32, FLOAT64 or BINARY, or left flexible with an asterisk. Sparse vectors are also supported for models that emit mostly zero values.
What to understand
Fixing the dimension count and format enforces consistency and lets the optimizer plan storage. FLOAT32 is the common default, INT8 saves space at some precision cost, and BINARY packs one bit per dimension for very compact storage with Hamming distance. Vectors sit in the same row as business columns, so ordinary SQL filters and joins still apply.
How it shows up
Expect a column definition to evaluate, or a question asking which format reduces storage for a large corpus.
Common confusion
Assuming vectors need a separate vector database; in Oracle they live in ordinary tables next to relational data.
1.3 Generate embeddings inside and outside the databaseONNX models in the database versus external providers.
What it means
Embeddings can be produced in the database by importing an ONNX-format model with DBMS_VECTOR.LOAD_ONNX_MODEL and calling VECTOR_EMBEDDING in SQL. They can also be produced by an external service, such as OCI Generative AI, Cohere or a local Ollama endpoint, through DBMS_VECTOR.UTL_TO_EMBEDDING or UTL_TO_EMBEDDINGS with a stored credential.
What to understand
In-database models avoid network calls and keep data private, which suits regulated data and batch jobs. External providers give access to larger hosted models but need a credential, network access control entries on non-Autonomous databases, and acceptance that text leaves the database. Oracle Machine Learning tooling can convert common sentence-transformer models to ONNX for import.
How it shows up
A scenario lists privacy, cost or model-size constraints and asks which embedding route satisfies them.
Common confusion
Mixing up LOAD_ONNX_MODEL, which imports a model, with UTL_TO_EMBEDDING, which calls a model to produce a vector.
1.4 Use vector SQL functions and constructorsTO_VECTOR, VECTOR_DIMENSION_COUNT, VECTOR_NORM and serialization.
What it means
Oracle supplies SQL functions to create and inspect vectors. TO_VECTOR (or the VECTOR constructor) builds a vector from a string or array literal, VECTOR_DIMENSION_COUNT and VECTOR_DIMENSION_FORMAT describe it, VECTOR_NORM returns its magnitude, and VECTOR_SERIALIZE or FROM_VECTOR turns it back into text.
What to understand
These functions matter when loading vectors created elsewhere, checking that a column matches a model's output, and debugging queries. Inserting a vector whose dimension count differs from a fixed column definition fails, so a quick VECTOR_DIMENSION_COUNT check is a common troubleshooting step before blaming the index or metric.
How it shows up
You may be asked which function confirms that an imported embedding has the expected number of dimensions.
Common confusion
Treating VECTOR_NORM as a distance; it measures one vector's length, not the gap between two vectors.

High-yield concepts

The terms, rules and formulas this domain turns into questions most often.

  • Vector embedding

    A numeric array that encodes the meaning of content.

    Remember
    Similar meaning means nearby vectors.
    Tested as
    Underpins every search, RAG and agent question.
    Trap
    Assuming any two models' vectors can be compared.

  • VECTOR data type

    Native column type in Oracle AI Database 26ai for embeddings.

    Remember
    Declared with dimensions and a format, or flexible with an asterisk.
    Tested as
    Table design questions.
    Trap
    Believing a separate vector store is required.

  • Vector formats

    INT8, FLOAT32, FLOAT64 and BINARY storage options.

    Remember
    BINARY is smallest and pairs with Hamming distance.
    Tested as
    Storage-saving scenarios.
    Trap
    Choosing FLOAT64 by habit when space matters.

  • Sparse vectors

    Vectors where most dimensions are zero, stored compactly.

    Remember
    Used with sparse models that capture keywords.
    Tested as
    Hybrid relevance scenarios.
    Trap
    Treating every embedding as dense.

  • ONNX in-database model

    An embedding model imported into the database in ONNX format.

    Remember
    Loaded with DBMS_VECTOR.LOAD_ONNX_MODEL and called by VECTOR_EMBEDDING.
    Tested as
    Privacy and latency scenarios.
    Trap
    Thinking the model runs in an external service.

  • VECTOR_EMBEDDING

    SQL function that applies an in-database model to input data.

    Remember
    Works in SELECT, INSERT and UPDATE statements.
    Tested as
    Pipeline and query questions.
    Trap
    Confusing it with UTL_TO_EMBEDDING, which can call outside providers.

  • UTL_TO_EMBEDDING(S)

    DBMS_VECTOR routines that create embeddings via a configured provider.

    Remember
    External providers need a credential and network access.
    Tested as
    Provider configuration questions.
    Trap
    Forgetting the credential step.

  • Same-model rule

    Query and stored vectors must come from one embedding model.

    Remember
    A model change means re-embedding the corpus.
    Tested as
    Troubleshooting poor results.
    Trap
    Blaming the index when the model changed.

  • TO_VECTOR and VECTOR_SERIALIZE

    Convert between text literals and vectors.

    Remember
    Useful for loading and inspecting data.
    Tested as
    SQL syntax questions.
    Trap
    Expecting implicit conversion in every context.

Side by side

Concepts candidates mix up, compared on the point that decides the answer.

Ways to produce embeddings

ConceptMeaningKey differenceCommon trap
In-database ONNX model Model imported and executed inside the database Data never leaves the database Forgetting the model must be ONNX format
External provider via DBMS_VECTOR Hosted model called with a credential Larger models, but network and data egress Skipping credential or ACL setup
Pre-computed vectors loaded Embeddings created elsewhere and inserted Database only stores and compares them Querying with a different model

Vector storage formats

ConceptMeaningKey differenceCommon trap
FLOAT32 32-bit floating point per dimension Common balance of precision and size Assuming it is the only choice
INT8 8-bit integer per dimension Smaller footprint with quantized precision Expecting identical ranking to FLOAT32
BINARY One bit per dimension Most compact, compared with Hamming distance Using cosine on binary vectors
Sparse Only non-zero dimensions stored Suits keyword-aware sparse models Declaring it for dense model output

Exam traps

Each trap is an answer that looks right. Learn the clue that should change your mind.

  1. Keyword search finds meaning

    What candidates pick
    Use LIKE or a text index to find paraphrased answers.
    Why it looks right
    Text search is familiar and already indexed.
    Why it's wrong
    Keyword matching needs shared terms; semantic search compares embeddings so paraphrases still rank highly.

    Clue to watch for The scenario stresses different wording with the same meaning.

  2. Same dimensions means compatible

    What candidates pick
    Compare vectors from two models because both output 768 dimensions.
    Why it looks right
    Matching sizes look interchangeable.
    Why it's wrong
    Each model defines its own vector space, so distances across models are meaningless even at equal size.

    Clue to watch for The question mentions a model upgrade or a second model.

  3. External call keeps data private

    What candidates pick
    Use a hosted provider when text must stay inside the database.
    Why it looks right
    The provider is an Oracle cloud service.
    Why it's wrong
    Any external provider receives the input text; only an in-database ONNX model keeps it inside the database boundary.

    Clue to watch for Words such as regulated, private network or no egress.

  4. FLOAT64 is always best

    What candidates pick
    Pick the most precise format for a very large corpus.
    Why it looks right
    More precision sounds safer.
    Why it's wrong
    Higher precision rarely changes ranking much but doubles storage; INT8 or BINARY fit space-constrained designs.

    Clue to watch for The scenario highlights storage cost or memory size.

  5. VECTOR_NORM measures similarity

    What candidates pick
    Use VECTOR_NORM to compare a query with stored rows.
    Why it looks right
    It is a vector function with a numeric result.
    Why it's wrong
    VECTOR_NORM returns a single vector's magnitude; similarity needs VECTOR_DISTANCE between two vectors.

    Clue to watch for The task compares two items rather than describing one.

Answer framework

Use this order when a foundation question asks how to store or create embeddings in Oracle AI Database.

  1. Decide whether the problem needs semantic similarity or exact matching, since only the first calls for embeddings.
  2. Identify the embedding model and its output, because dimension count and format drive the VECTOR column definition, for example FLOAT32 with 384 dimensions for a small sentence model or BINARY for packed embeddings.
  3. Check privacy and network constraints to choose between an imported ONNX model and an external provider with a credential; an ONNX model loaded with DBMS_VECTOR.LOAD_ONNX_MODEL keeps text inside the database.
  4. Pick the SQL function that fits the step: VECTOR_EMBEDDING to create, TO_VECTOR to convert, VECTOR_DIMENSION_COUNT to inspect.
  5. Confirm that queries will embed search text with the same model used for stored rows, otherwise rankings will be meaningless.

Practice question

A healthcare provider must embed patient notes stored in Oracle AI Database 26ai. Policy forbids sending note text to any service outside the database. Which approach meets the requirement?

Want more like this? The 1Z0-1159-26 practice test has questions for every domain.

Checklist

0 of 11

Tick each one you can do without notes. Your ticks stay in this browser.

Study strategy

Start hands-on, because this section is easier to remember after you have typed the SQL. Use a free Autonomous AI Database or an Oracle AI Database 26ai Free container and create a table with a VECTOR column. Insert a few vectors with TO_VECTOR, then read them back with VECTOR_SERIALIZE and VECTOR_DIMENSION_COUNT.

Next, import a small ONNX sentence-transformer model with DBMS_VECTOR.LOAD_ONNX_MODEL and embed a dozen sentences with VECTOR_EMBEDDING. Then configure one external provider with a credential and call UTL_TO_EMBEDDING, so you can explain the trade-off between the two routes from experience.

Write a one-page sheet listing each vector format, its size per dimension and its matching distance metric, and add the same-model rule in bold. Finish with short practice questions on column declarations and embedding routes. Keep sessions brief: this domain is the base layer, and most of your remaining hours belong to indexing, RAG and agents, where these definitions get applied under pressure. Before moving on, run a quick self-test: explain from memory why a model upgrade forces a full re-embedding of stored rows.

FAQ

How much of 1Z0-1159-26 is Domain 1, Explain Oracle AI Vector Search Foundations?

Oracle does not publish section weights for this exam. Treat this foundations section as a smaller block whose ideas reappear in every later domain.

Do I need a separate vector database with Oracle?

No. Oracle AI Database 26ai stores embeddings in a native VECTOR column inside ordinary tables, so vectors can be filtered and joined with business data.

What is the difference between VECTOR_EMBEDDING and UTL_TO_EMBEDDING?

VECTOR_EMBEDDING applies an ONNX model already imported into the database. DBMS_VECTOR.UTL_TO_EMBEDDING can call an external provider, such as OCI Generative AI, through a stored credential.

Which vector format saves the most space?

BINARY stores one bit per dimension and is compared with Hamming distance. INT8 is a middle option that keeps more precision than BINARY.

Can I compare embeddings from two different models?

No. Each model has its own vector space. If you switch models you must re-embed the stored rows before similarity scores mean anything.

Did this exam replace Oracle AI Vector Search Professional (1Z0-184)?

Our catalog lists it that way, but we could not find an Oracle statement confirming the replacement. Check the Oracle University exam catalog for the current status of both codes.

Which product version should I practice on?

Use Oracle AI Database 26ai, either the Free edition or an Autonomous AI Database, since vector search, Select AI and agent features are documented there.

Other Artificial Intelligence & Machine Learning certifications

Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.

Test your Domain 1 knowledge

You have covered embeddings, the VECTOR data type, in-database and external embedding models and the vector helper functions. Check whether you can choose the right setup for a scenario.

Take the 1Z0-1159-26 practice test Check your 1Z0-1159-26 readiness

Keep going