Database Administration
Database administration has become a control point for AI systems: production stores now hold the embeddings and governed records that retrieval-augmented applications and model training depend on, while schemas, queries, and migrations are increasingly drafted by AI coding agents and executed by automated service accounts. The course therefore adds vector and hybrid indexing, benchmarking of AI-assisted and learned query optimization, provenance for training data, and least-privilege access and auditing for agent traffic, without displacing core relational design and operations.
Current description → proposed description
This course provides students with solid theoretical and practical knowledge for developing database management systems. Students will plan, design, implement, maintain, and use database management systems and review the use of databases in small and large commercial organizations. The course addresses concepts, database structures, database architecture, understanding user requirements, user views, functions, and evaluation of database management systems. The course focuses on the relational database model, standard SQL language, database structure normalization, conceptual data modeling, and the entity-relationship data model. Students will work with real world applications and databases. Concepts of data integrity, security, privacy, ethical use, and concurrence control are included.
This course develops the theory and practice of designing, implementing, and administering database systems that serve production applications and data-intensive analytics. Students plan and build databases from ambiguous requirements, working through conceptual data modeling, the entity-relationship model, normalization, standard SQL, physical design and indexing, transaction isolation and concurrency control, backup and recovery, and the evaluation of competing database management systems in small and large organizations. Coursework extends these foundations to the demands contemporary AI practice places on the data tier: storing and indexing vector embeddings alongside relational data for retrieval-augmented enterprise applications, tuning hybrid retrieval for recall, latency, and cost, and maintaining reproducible, reversible load and migration paths for the data that feeds analytics and model training. Because schemas, queries, and migrations are now routinely drafted by AI coding agents and executed by automated services, students learn to review machine-generated SQL, benchmark AI-assisted and learned query optimization against measured baselines, and constrain agent and service accounts through least-privilege roles, row-level security, masking, and auditing. Throughout, the course treats data integrity, security, privacy, retention, provenance, and the ethical reuse of institutional data as central administrative responsibilities.
What changes
- Vector and hybrid indexing for retrieval-augmented enterprise applications
- Review and benchmarking of AI-generated SQL and learned query optimization
- Least-privilege roles, row-level security, and auditing for AI agent and service accounts
- Provenance and reproducibility for data reused in analytics and model training
- Graduate framing around design judgement and measured trade-offs rather than SQL practice
5 proposed outcomes, mapped to 5 program outcomes
Each outcome below is written to be observable and assessable, and each is mapped to the program learning outcomes for which it produces evidence.
Students will be able to design a normalized relational schema together with its physical and operational implementation from ambiguous organizational requirements, justifying each normalization, denormalization, indexing, isolation-level, concurrency-control, and recovery-objective decision against the stated workload, including high-volume automated and agent-issued traffic, the governing integrity and consistency requirements, and projected growth.
PLO 2.1 asks students to conduct "storing ... data ... using ... databases ... to solve real-world problems," and the evidence for it here is the conceptual-to-physical storage design itself, deriving a schema, its normalization, and its indexing from ambiguous organizational requirements; the isolation, concurrency, and recovery settings are included as parts of the same design artifact and are not offered as the outcome evidence.
Students will be able to construct the load and migration path for a production database, covering bulk and incremental load design, enforcement of integrity and referential constraints at load time with defined handling for rejected and quarantined records, and online schema migration deployed reversibly and under version control against a live workload.
Load-time enforcement of integrity and referential constraints, with defined handling for rejected and quarantined records, is the "cleaning" clause of PLO 2.1, and designing the bulk, incremental, and migration paths by which those records are landed is the "storing ... using ... databases" clause; the CLO makes no claim on the outcome's "collecting" or "APIs" language, which belongs to CSC 6240.
Students will be able to critique AI-assisted query and schema optimization, including machine-generated SQL, index and migration recommendations, and learned cardinality or plan-selection models, by benchmarking proposed changes against measured execution plans and baseline workload metrics and documenting the conditions under which the recommendations degrade performance or correctness.
Benchmarking learned cardinality and plan-selection models against execution plans and baseline workload metrics, then documenting where they fail, is precisely the assessment of a model's performance and limitations with appropriate metrics and diagnostic-driven refinement that PLO 6.2 requires.
Students will be able to implement vector and hybrid retrieval within a production data store, spanning embedding storage, approximate-nearest-neighbor index selection and parameterization, metadata filtering, and chunk modeling, tuned to meet an enterprise retrieval-augmented application's stated recall, latency, and cost targets.
PLO 6.1 calls for using LLMs in industrial settings through embeddings and structured deployment in enterprise environments, and this CLO's embedding storage, index parameterization, and tuning against an enterprise application's service-level targets are the data-tier form of that deployment work.
Students will be able to justify a governance and access-control design for a database serving AI workloads, addressing least-privilege roles for human and agent service accounts, row-level security, masking and de-identification, retention and consent handling, audit trails for machine-issued statements, and provenance records for data reused to train or evaluate models.
The privacy controls, accountability through audit trails for machine-issued statements, and provenance for data reused in training are the ethical-and-legal-integrity evidence for this outcome, which the workbook states twice in word-for-word identical language, once under Christian Faith as PLO 1.1 and again under Integrated Disciplinary Knowledge as PLO 3.2, so a CLO that earns one necessarily earns the other.
Program outcomes this course reaches
Filled cells are program learning outcomes with at least one supporting course learning outcome in this course. Sparse coverage is expected — no single course carries all twelve.