CSC 6240 3 Credits 6000 level Substantial AI weight MSAI core

Data Acquisition and Management

Agentic systems have made data infrastructure the substrate of AI behavior: retrieval corpora, embeddings, permission-aware access and provenance now determine what an agent can know, do and be held accountable for, and ingested content has become an attack surface through poisoning and indirect prompt injection. The course therefore extends classic acquisition, ETL and governance work to agent-facing retrieval layers, dataset lineage and licensing, and the evaluation of pipelines and pipeline code produced with AI coding agents.

The revision

Current description → proposed description

Current — CUW catalogverbatim

This course provides a comprehensive foundation in acquiring, managing, and preparing data for use in intelligent systems and data-driven applications. Students will explore the full data lifecycle, from sourcing and ingestion to cleaning, transformation, storage, and retrieval within the context of real-world technical and ethical challenges. Topics include web scraping, API integration, database design, data warehousing, ETL pipelines, data governance, and schema evolution. Students will also address data quality assurance, handling unstructured data, version control, and compliance with legal and ethical standards related to privacy and ownership. Special attention is given to managing large-scale, multi-source datasets in AI and machine learning workflows. By the end of the course, students will be able to design robust data infrastructures that supports scalable AI applications.

Proposed — revised for the AI eradraft

This course provides a comprehensive foundation in acquiring, managing, and preparing data for use in intelligent systems, data-driven applications, and autonomous AI agents. Students will explore the full data lifecycle, from sourcing and ingestion to cleaning, transformation, storage, and retrieval within the context of real-world technical and ethical challenges. Topics include web scraping, API integration, database design, data warehousing, ETL and ELT pipelines, data governance, and schema evolution, extended to the retrieval infrastructure that agentic systems depend on: chunking and embedding strategies, vector and hybrid indexes, permission-aware retrieval, and data services exposed to tool-calling agents through protocols such as the Model Context Protocol. Students will also address data quality assurance, handling unstructured data, version control, lineage and provenance, and compliance with legal and ethical standards related to privacy and ownership. Special attention is given to managing large-scale, multi-source datasets in AI and machine learning workflows, to the risks that ingested content introduces including poisoning, indirect prompt injection, and license contamination, and to the evaluation and review of pipeline code produced with AI coding agents. By the end of the course, students will be able to design robust, observable data infrastructures that support scalable AI applications.

Note. PLO 1.1 and PLO 3.2 are word-for-word identical in the workbook and carry identical I/D/AE rows for this course (both "I"); CLO 6 is mapped to both as instructed, and no distinction between them is invented. The catalog description for CSC 6240 is already word-for-word the department's workbook proposal, so desc_source is "generated" but the text above is an additive evolution of adopted department work - the original's opening scope, topic list and closing design commitment are preserved, with agentic retrieval infrastructure, provenance, ingestion-borne risk and pipeline evaluation added; "infrastructures that supports" in the current catalog text is corrected to "infrastructures that support". The CLO set covers all eight PLOs the workbook's curriculum map lists for CSC 6240 and adds none beyond them. All five reported problems were valid on inspection of the source files and all five are fixed: CLO 1's relative clause now carries the acquisition through to a downstream machine-learning analysis of a real-world question, closing the exit-assessed half of PLO 2.1 that the prior CLOs left uncovered; CLO 2's rationale no longer presents an elided string as verbatim workbook text and instead paraphrases PLO 6.1, whose actual wording is "structured deployment in embedded systems and enterprise environments"; "stratified sampling" is removed from both CLO 3 and its rationale and replaced with blocking and candidate generation, leaving only probabilistic record matching and vector similarity attributed to PLO 3.1's named probabilistic-model and search-algorithm foundations; CLO 4 now validates each dataset by training baseline supervised and unsupervised models on it and narrows the reinforcement-learning claim to preference and trajectory data for offline RL or RLHF, with the rationale kept proportional to PLO 4.1's "Introduced" row; and CLO 5 now evaluates both the managed data assets and the retrieval-augmented system's answers, with groundedness, citation accuracy and hallucination rate named in the rationale as the model-facing evidence for PLO 6.2 and refinement targeted at chunking, embedding and index configuration rather than "the pipeline". Nothing else was changed.

What changes

Retrieval corpora: chunking, embeddings, vector/hybrid indexes
Agent-facing data services and permission-aware access (MCP)
Provenance, lineage and licensing of training corpora
Ingestion-borne risk: poisoning and indirect prompt injection
Evaluation of pipelines and agent-authored pipeline code

Course learning outcomes

6 proposed outcomes, mapped to 8 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 an end-to-end data acquisition and management architecture that ingests, reconciles, stores, and serves heterogeneous data from web sources, vendor APIs, event streams, and operational databases to a downstream machine-learning analysis of a real-world question, within explicit licensing, rate-limit, cost, and freshness budgets.

Maps to

PLO 2.1 Data Acquisition for Real Problems

The architecture is built from the exact sources PLO 2.1 names - APIs, databases and multi-source feeds - and carries them through collecting, cleaning and storing all the way to the machine-learning analysis of a real-world question that completes the outcome, which is the end-to-end acquisition competency the workbook assesses at exit in this course.

Students will be able to construct governed retrieval infrastructure for enterprise corpora - curation, chunking, embedding, and hybrid vector and keyword indexing - exposed to tool-calling LLM agents through permission-aware interfaces such as Model Context Protocol servers.

Maps to

PLO 6.1 LLMs in Industrial Settings

Embedding and indexing enterprise corpora and serving them to LLM agents over governed interfaces is precisely the use of APIs, embeddings and structured enterprise deployment that PLO 6.1 names, at the introductory level the workbook assigns this course.

Students will be able to implement entity resolution, deduplication, and schema-evolution strategies across multi-source datasets using probabilistic record matching, blocking and candidate generation, and vector-similarity methods appropriate to the error and cost tolerances of the target application.

Maps to

PLO 3.1 AI Theory and Mathematics PLO 2.1 Data Acquisition for Real Problems

Probabilistic record matching and vector-similarity search are the probabilistic-model and search-algorithm foundations PLO 3.1 names, applied here - with blocking and candidate generation to keep the matching tractable - to the cleaning and reconciliation of multi-source data that PLO 2.1 requires.

Students will be able to construct versioned, reproducible feature and training datasets - each validated by training baseline supervised and unsupervised models on it, and, where the target is an agent, extended to preference and trajectory data suitable for offline reinforcement learning or RLHF - with documented lineage, disciplined splits, and labeling workflows in which labels proposed by AI agents are adjudicated by humans.

Maps to

PLO 4.1 Learning-Based Systems and Agents PLO 2.1 Data Acquisition for Real Problems

Training baseline supervised and unsupervised models on each dataset, and preparing preference and trajectory data for offline reinforcement learning or RLHF, is the introductory hands-on development of learning-based systems and agents that PLO 4.1 marks as Introduced here, while the versioned, lineage-documented, human-adjudicated production of that data is PLO 2.1 evidence.

Students will be able to evaluate managed data assets and the answers of the retrieval-augmented system built on them - completeness and conformance checks, drift monitors, and retrieval precision and recall for the assets; groundedness, citation accuracy, and hallucination rate for the answers; and the correctness of agent-authored pipeline code - with the resulting diagnostics driving successive refinement of chunking, embedding, and index configuration.

Maps to

PLO 6.2 Model Evaluation and Refinement

Groundedness, citation accuracy and hallucination rate scored on the system's own answers are direct evidence of the performance and limitations of an AI model that PLO 6.2 requires students to assess, the completeness, conformance, retrieval and drift measures supply its appropriate evaluation metrics, and the CLO makes those diagnostics drive iterative refinement of the model-backed retrieval configuration rather than stopping at measurement.

Students will be able to justify a data stewardship and governance plan for an agent-facing data platform - covering consent, ownership, licensing and provenance of ingested and training corpora, PII minimization, permission-aware retrieval, and defenses against poisoning and indirect prompt injection - in a dataset datasheet and data contract written for both engineering and non-technical stakeholders.

Maps to

PLO 1.1 Ethical and Legal Integrity in AI PLO 3.2 Ethical and Legal Integrity in AI PLO 5.1 Communicating AI to Any Audience

Consent, ownership, licensing, provenance and PII minimization are the privacy, accountability and potential-harm concerns of the ethical-and-legal-integrity outcome, which the workbook states twice in identical words - as PLO 1.1 under Christian Faith and as PLO 3.2 under Integrated Disciplinary Knowledge - with identical I/D/AE rows, so this CLO supports both; the required datasheet and data contract written for both engineering and non-technical stakeholders is the mixed-audience communication of results and ethical considerations named in PLO 5.1.

Coverage

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.

ULO1

1.11.2

ULO2

2.12.2

ULO3

3.13.2

ULO4

4.14.2

ULO5

5.15.2

ULO6

6.16.2

The workbook maps this course too

The proposal workbook's assessment map already assigns program learning outcomes to this course. The outcomes above were written to cover it.

Program outcome	Workbook level	In this draft
PLO 1.1 Ethical and Legal Integrity in AI	I	covered
PLO 2.1 Data Acquisition for Real Problems	D,AE	covered
PLO 3.1 AI Theory and Mathematics	I	covered
PLO 3.2 Ethical and Legal Integrity in AI	I	covered
PLO 4.1 Learning-Based Systems and Agents	I	covered
PLO 5.1 Communicating AI to Any Audience	I	covered
PLO 6.1 LLMs in Industrial Settings	I	covered
PLO 6.2 Model Evaluation and Refinement	I	covered

I = Introduced · D = Developed · AE = Assessed at Exit