DP-800
Developing AI-Enabled Database Solutions
DP-800 Exam Guide: Developing AI-Enabled Database Solutions
DP-800 validates whether you can design, secure, optimize, deploy, and extend AI-enabled database solutions across SQL Server, Azure SQL, and SQL databases in Microsoft Fabric. It serves database developers and data professionals who already work with T-SQL and need to connect database engineering with embeddings, vectors, models, and GitHub CI/CD. This guide helps you decide whether your experience is ready for the exam, which skill areas deserve the most practice, and how to build a focused study plan instead of collecting disconnected resources.
What does DP-800 validate?
DP-800 tests practical design and development judgment across Microsoft SQL platforms, not isolated familiarity with AI terminology. The target role combines database implementation, application integration, security, performance, deployment, and AI capabilities in solutions that use structured and semi-structured data.
Microsoft describes the target candidate as someone with subject matter expertise across Microsoft SQL Server, Azure SQL, and SQL databases in Microsoft Fabric. The profile also expects T-SQL database-development experience, familiarity with GitHub continuous integration and continuous deployment practices, and working knowledge of AI-assisted development tools, embeddings, vectors, and models.
The responsibilities extend beyond writing queries. They include designing database solutions, integrating AI features into scalable enterprise applications, securing and optimizing solutions, and implementing AI capabilities in databases. The role commonly intersects with application developers, DBAs, architects, AI engineers, DevSecOps engineers, and security and compliance administrators.
That combination matters when you decide how to prepare. A candidate who knows T-SQL but has not built deployment pipelines or vector search should not treat DP-800 as another conventional SQL exam. Conversely, an AI developer who has used embeddings but cannot reason about constraints, permissions, execution plans, or schema deployment has a different gap to close.
Who is the intended candidate?
The certification is classified as Intermediate, with Azure as the product, Developer as the role, and Data management as the subject. The associated course describes its audience as data professionals building AI-enabled database solutions across SQL Server, Azure SQL, and SQL databases in Microsoft Fabric.
Use that profile as a readiness test rather than a strict prerequisite list. Microsoft’s published material describes expected experience and skills; it does not establish a separate formal prerequisite for sitting the exam.
How are the DP-800 skills weighted?
The current blueprint assigns 35–40% to Design and develop database solutions, 35–40% to Secure, optimize, and deploy database solutions, and 25–30% to Implement AI capabilities in database solutions. Plan around the named domains, because the two largest domains together cover most of the assessment.
The study guide states that these skills are measured as of March 12, 2026. Blueprint percentages describe the relative emphasis of the exam, not a guaranteed number of questions. Related topics may also appear, and the bullets under each skill are examples of how Microsoft assesses the domain.
A sensible allocation is to give the first two domains equal priority, then use the AI domain as a focused specialization rather than leaving it until the final study session. Your personal time should still follow your gaps: a strong database developer may need more AI and CI/CD practice, while an AI-focused developer may need more security and performance work.
Design and develop database solutions: 35–40%
This domain covers database objects, advanced T-SQL, structured and semi-structured data, and AI-assisted SQL development. The published topics include tables, data types, indexes, columnstore indexes, specialized tables, JSON, constraints, sequences, partitioning, views, functions, stored procedures, triggers, common table expressions, window functions, error handling, graph queries, regular expressions, and fuzzy string matching.
Study by implementing small database features rather than reading every item as a vocabulary list. For example, create a schema that uses constraints, JSON data, a programmable object, a windowed query, and a performance-relevant index. Then explain why each design choice fits the requirement and what trade-off it introduces.
AI-assisted development is part of this domain as well. Practice reviewing generated SQL for security, correctness, data exposure, and maintainability. The goal is not to accept a tool’s answer automatically; it is to recognize when generated code conflicts with the data model or operational requirement.
Secure, optimize, and deploy database solutions: 35–40%
This domain joins three activities that are often studied separately: protecting data, diagnosing performance, and delivering controlled changes. The blueprint includes encryption, dynamic data masking, row-level security, object permissions, passwordless access, auditing, managed identities, endpoint protection, transaction isolation, concurrency control, execution plans, DMVs, Query Store, blocking, deadlocks, SQL database projects, testing, source control, schema drift, deployment pipelines, Azure integration, Data API builder, monitoring, and change-processing options.
Build a single practice scenario that moves from development to deployment. Define who can read which rows, choose an appropriate protection mechanism, inspect a slow query, and package a schema change in source control. Add a test and an approval point before deployment. This sequence forces you to connect the security, optimization, and delivery decisions that the domain emphasizes.
Do not study CI/CD as GitHub terminology alone. Be able to reason about branch strategy, pull requests, conflict resolution, secrets, authentication, code ownership, approvals, reference data, schema validation, and drift detection. Also distinguish change tracking, CDC, change event streaming, Azure Functions with SQL trigger bindings, and Logic Apps by the type of change-processing requirement they address.
Implement AI capabilities in database solutions: 25–30%
This domain covers the path from source data to useful AI retrieval: models and embeddings, intelligent search, and retrieval-augmented generation. The blueprint includes evaluating external models, creating and managing them, selecting an embedding-maintenance method, choosing columns for embeddings, chunking content, generating embeddings, implementing full-text, semantic vector, and hybrid search, using vector indexes and metrics, applying reciprocal rank fusion, evaluating search performance, and building RAG flows.
A practical lab should use a small document or product dataset. Decide which columns belong in the embedding, split content into meaningful chunks, store vectors with the required metadata, and compare full-text, vector, and hybrid retrieval. Then trace how retrieved content becomes JSON for a language model and how the model response is extracted.
Learn the decision logic behind the features. Vector search is not automatically better than full-text search; hybrid retrieval is not automatically necessary; and a larger model is not automatically the correct choice. Consider query intent, language, structured filters, latency, index behavior, and the quality of the returned context.
Which experience gaps should you close first?
Start with a skills inventory mapped to the three official domains. Mark each topic as can implement, can explain, or unfamiliar. Schedule hands-on work for unfamiliar items and explanation practice for topics you can use only by following a recipe.
For the database-development domain, check whether you can design constraints and indexes, write advanced T-SQL, work with JSON and semi-structured data, and choose among specialized table or indexing options. For the security and delivery domain, check whether you can explain access boundaries, diagnose blocking or deadlocks, and move a schema through source control and deployment controls. For the AI domain, check whether you can design an embedding pipeline, select a search strategy, and describe a RAG data flow.
Do not let broad job titles hide narrow gaps. A person who administers Azure SQL every day may still need deliberate practice with graph queries or AI-assisted SQL development. A data scientist may understand embeddings but need to build permissions, audit controls, deployment validation, and performance investigations.
Use the inventory to choose your study route. If most topics are unfamiliar, follow the official learning sequence before attempting repeated practice questions. If the core SQL and delivery skills are established, use the blueprint to target the AI topics and any platform-specific areas you have not implemented.
A useful readiness check
You are closer to exam readiness when you can take a scenario, identify the governing requirement, select a Microsoft SQL capability, implement a basic version, and explain how you would secure, test, monitor, and deploy it. Memorizing product names without making those connections is a weak signal.
For each domain, produce a short decision record. State the requirement, selected approach, rejected alternative, security or operational consequence, and test you would run. This format exposes shallow understanding quickly and gives you reusable revision notes.
What should you build while studying?
Build one compact portfolio project and extend it through the blueprint instead of creating unrelated demonstrations. A catalog, support-knowledge base, or internal-document dataset can support structured tables, semi-structured content, search, embeddings, security, performance testing, and deployment exercises.
Begin with a relational model containing keys, foreign keys, checks, defaults, and indexes. Add JSON content where it makes sense, then create views, procedures, functions, triggers, CTEs, and window-function queries. Keep the schema small enough that you can change it repeatedly and inspect the consequences.
Next, add data protection and operational controls. Apply row-level or object-level access according to a stated user requirement, configure auditing in the platform available to you, and test how permissions affect queries. Capture a deliberately inefficient query, inspect its execution behavior, and use Query Store or relevant diagnostic views where available in your environment.
Put the schema under source control through a SQL database project workflow. Practice a branch, a pull request, a schema validation step, a test, secret handling, and an approval-controlled deployment. Introduce a schema change that creates drift, detect it, and then reconcile the project and target.
Finish with AI retrieval. Select embedding columns, define chunk boundaries, generate vectors, test full-text and vector retrieval, and add a hybrid path. Represent retrieved structured data as JSON before sending it to a language model. Document why your search and maintenance choices fit the workload.
The project is a study instrument, not evidence that the live exam will reproduce the same scenario. Its value comes from making you perform and defend the decisions represented in the blueprint.
How to keep the project efficient
Use one dataset and change one design variable at a time. If you alter the schema, access policy, index, embedding method, and deployment process simultaneously, you will not know which change caused a result. Keep a decision log and record the query, configuration, expected behavior, observed behavior, and correction.
Work from official documentation linked by the study guide when a feature is new or platform-specific. Microsoft notes that most questions cover General Availability features, although commonly used Preview features may also appear. Check the current study guide before final revision rather than relying on an old course or third-party summary.
How should you sequence the official preparation resources?
Use the DP-800 study guide as the control document, then use Microsoft’s preparation paths and course to fill specific gaps. The official preparation content is organized as Design and develop database solutions, Secure, optimize, and deploy database solutions, and Implement AI capabilities in database solutions.
A productive sequence is foundations first, operational delivery second, and AI integration third. This mirrors how a real solution is built: establish the data model and code, protect and operate it, then add retrieval and model interactions. You can reverse the order if your current role makes one domain dominant, but do not postpone the two heavier domains because the AI material feels newer or more interesting.
Microsoft also lists the instructor-led DP-800T00-A course, Develop AI-enabled database solutions, as Intermediate and describes it as a 3-day course. The course is available for instructor-led or self-directed learning and is listed in English, Chinese Simplified, Chinese Traditional, French, German, Italian, Japanese, Korean, Portuguese Brazil, and Spanish. Treat the course as structured instruction, not a substitute for implementing the skills.
The certification page provides a Practice Assessment through AI Skills Navigator, and sign-in is required to launch it. Use it after your first implementation pass, not as your only study resource. Microsoft says Practice Assessments are free and can be taken repeatedly; review why an answer is correct and identify the missing concept instead of recording only a score.
The exam sandbox is another official resource. Use it to learn the interface and question interaction before scheduling or sitting the assessment. It cannot teach database design or vector search, but it can reduce avoidable uncertainty about how you will work through the assessment environment.
How to use third-party material safely
Use third-party books, courses, or practice tests only as supplementary explanations and verify their coverage against the current Microsoft study guide. Microsoft states that it does not review third-party study materials and is not responsible for keeping them aligned with product updates.
Avoid any resource claiming that memorizing recalled or leaked questions guarantees a pass. Such material does not build the implementation judgment DP-800 describes and can leave major blueprint areas unprepared.
What is a practical DP-800 study roadmap?
A four-stage roadmap works well when you already have some SQL experience: map the blueprint, build the database core, operationalize it, and add AI retrieval before validating readiness. Adjust the calendar to your own schedule; the stages matter more than an arbitrary number of study days.
Stage one is a diagnostic pass. Read the current study guide, list every topic under its domain, and label your experience honestly. Run a short implementation exercise in each domain. Your output should be a gap list with concrete actions such as implement row-level security, investigate a deadlock, create a SQL database project, or compare vector and hybrid search.
Stage two is database construction. Create the relational and semi-structured model, then implement database objects and advanced T-SQL. Include constraints, indexing, JSON, programmable objects, error handling, and at least one less-familiar query style. For every feature, write a brief note explaining when you would use it and what could go wrong.
Stage three is security, performance, and delivery. Add permissions, encryption or masking decisions, auditing, transaction and concurrency reasoning, query diagnosis, source control, testing, schema validation, secrets, approvals, and deployment. Rebuild the project from source rather than relying on manual changes. This is where conceptual familiarity becomes operational confidence.
Stage four is AI integration and exam validation. Implement embeddings, chunking, model evaluation, vector search, hybrid search, and a RAG flow. Then take the official Practice Assessment, review the sandbox, and return to the blueprint for any weak domain. Schedule only when you can explain and reproduce the core workflows without step-by-step copying.
Keep the final review selective. Revisit decision points, platform terminology, security consequences, performance evidence, and deployment controls. Do not spend the last session trying to memorize every function name while neglecting the architecture and troubleshooting reasoning that connects the skills.
What should each study session produce?
Each session should leave a tangible artifact: working SQL, a diagram, a query plan observation, a security test, a pipeline definition, a search comparison, or a decision note. If a session produces only highlighted text, follow it with a hands-on exercise before moving on.
At the end of the roadmap, create a one-page gap register. For every remaining weakness, record the official domain, the feature or decision involved, the documentation to review, and the smallest exercise that will prove understanding. Stop adding new resources when the register is more useful than another reading list.
What delivery details should you confirm before scheduling?
Microsoft states that DP-800 is proctored and provides 120 minutes for the assessment. The certification page says the exam may include interactive components, while Microsoft’s FAQ explains that lab delivery depends on connectivity and technical requirements and is currently limited to qualifying test centers.
Do not assume that every candidate will see identical lab availability. Microsoft says the presence or absence of labs can change based on factors such as Azure availability, outages, and bandwidth. When labs are included, the exam interface indicates the number of items, case studies, and labs, including task counts, when you launch the exam.
If a lab opens to a blank white screen, Microsoft’s FAQ suggests opening a new tab and navigating to https://portal.azure.com. This is a technical recovery instruction, not a reason to depend on a particular test-day experience. If you have a specific testing center in mind, confirm whether it can deliver labs; eligible centers are presented during scheduling.
The exam is currently listed as offered in English on the certification page. Microsoft explains that some exams are localized and that localized versions are updated approximately eight weeks after the English version. If the exam is unavailable in your preferred language, the study guide says you can request an additional 30 minutes. Arrange any language or disability accommodation in advance through Microsoft’s stated process.
You can schedule Microsoft certification exams through Pearson VUE no more than 90 days in advance, and Microsoft’s FAQ says a maximum of two certification exams can be scheduled at one time through Pearson VUE. Register with a personal Microsoft account, as the DP-800 certification page strongly recommends it to preserve your exam records.
How much does DP-800 cost?
Confirm the current price on the DP-800 exam page before scheduling. Microsoft says exam prices vary by country or region, may have additional taxes, and are subject to change; academic pricing and special offers may also apply. Avoid treating a general Microsoft exam price as a guaranteed DP-800 charge.
What happens after an unsuccessful attempt?
Microsoft’s DP-800 certification page says a failed attempt can be retaken 24 hours after the first attempt, while later retake timing varies. Use the score report and your gap register to choose the next study action rather than immediately repeating the same practice routine.
What mistakes undermine DP-800 preparation?
The most damaging mistake is studying the AI vocabulary separately from database engineering. DP-800 connects embeddings and search with data modeling, security, performance, maintenance, and application integration. Your exercises should make those connections explicit.
Another mistake is treating the percentage ranges as a checklist of guaranteed question counts. The official blueprint gives domain emphasis and illustrative topics, not a script for the exam. Prepare the underlying decisions and related concepts rather than trying to predict the item mix.
Candidates also lose value by reading implementation lists without building anything. A page about row-level security is not equivalent to testing two users against protected rows. A description of vector search is not equivalent to comparing retrieval results, index choices, and filtering behavior.
Do not ignore deployment because the title sounds database-focused. SQL database projects, source control, schema drift, testing, secret management, approvals, and deployment controls are explicitly included in the security, optimization, and deployment domain.
Finally, do not use practice assessments as a memorization loop. Repeatedly selecting answers until a score improves can conceal the same conceptual gap. For every missed item, explain the requirement, the selected capability, the rejected alternative, and the operational consequence in your own notes.
A final error check
Before scheduling, ask whether you can complete a small end-to-end change: modify a schema, validate it, test access, inspect performance, deploy it through controlled steps, update dependent embeddings, and verify retrieval. You do not need a production-sized system, but you do need to understand the chain of decisions.
Also confirm that your notes are tied to the current study guide. Microsoft can update exam skills and associated technologies. Review the official page again shortly before the exam and replace obsolete notes rather than adding more unverified summaries.
What should you do next?
Open the current DP-800 study guide and turn its three named domains into a personal checklist. Then choose one small database project that can demonstrate T-SQL, security, performance, deployment, embeddings, search, and RAG decisions. Schedule after your implementation evidence and Practice Assessment review support readiness, not simply because you have finished reading.
If your checklist exposes a weak foundation in SQL or database delivery, follow the relevant official learning path or the DP-800T00-A course before concentrating on advanced AI features. If the foundation is strong, spend the remaining preparation time on the AI workflow, platform differences, and end-to-end integration.
Before booking, verify the current exam page for language, price, scheduling availability, accommodations, and delivery conditions. Use a personal Microsoft account, explore the exam sandbox, and confirm the testing center information shown through Pearson VUE. Keep the study guide as your final authority when a third-party resource conflicts with it.
The objective is not to predict the exam’s questions. It is to become able to choose, implement, secure, troubleshoot, and deploy the database and AI capabilities represented by the blueprint. That is the preparation standard most likely to remain useful when Microsoft updates the underlying services.
Conclusion
DP-800 preparation is strongest when it follows the work of delivering an AI-enabled database: model the data, write and test the SQL, secure access, investigate performance, control deployment, maintain embeddings, and evaluate retrieval. Use the official blueprint to find gaps, build evidence through small implementations, and confirm scheduling details directly with Microsoft before committing to an exam appointment.