Kamin capstone team plan

Starting point

The baseline includes a local profile, transcript review, an explainable knowledge graph, SPARQL, RDF/CSV exports, optional browser-based semantic recall and encrypted capability snapshots. The baseline also includes a short starter tour, 9 career and 10 practice pathways, and a proposed institutional and field-validation roadmap. Record the starting release and distinguish new student contributions. The three-college field study is planned; no field results are claimed.

From a prototype to an AI-native product

The baseline now includes six practical tasks, saved progress and local reviews, pathway comparison, project capability suggestions and bounded credential import. These are inherited features; record new contributions separately.

• Divide work into isolated units with an outcome, inputs/outputs, constraints and acceptance criteria written before implementation.
• Every change has a branch, a different human reviewer, reproducible evidence and rollback. Agents support execution and testing; a named human remains accountable.
• Build one vertical slice on staging with synthetic data: identity and permissions, an API, a database and separately consented optional storage. These services are planned, not active in the public release.
• The Render static site is real hosting. Next: an approved domain, staging/production separation, monitoring, backup restoration and an operational owner.
• Challenge user isolation, sharing withdrawal, unsupported claims and prompt injection. Measure time to informed acceptance/rejection and confirmed defects escaping review.

AI-native product delivery plan · Task contract and acceptance evidence template

Leadership responsibilities

• Student 1: student journeys, Arabic/English, mobile, accessibility, user guidance and deployment.
• Student 2: ontology and RDF, sourced relationships and rules, SPARQL competency questions and SHACL validation.
• Student 3: evaluate and improve local semantic recall, deliver one explainable recommendation service, baselines and data-quality tests.

Integration, testing and privacy belong to everyone. Every change has an implementer and a peer reviewer.

First two weeks

• Run the project on all three machines; record the starting commit and test results.
• Exercise six tasks: profile without a transcript, synthetic evidence review, recommendation explanation, save/restore, deletion and graph query.
• Prioritize ten improvements with acceptance criteria and ten competency questions with expected answers.
• Submit one small peer-reviewed pull request per student and present a live team demo.
• Start coordination with three colleges and two HR managers the supervisor can introduce. Participation is not yet confirmed.

End-of-semester release gate

• Weeks 1–4: baseline, early protocol review and coordination, limited sourced coverage for each college.
• Weeks 5–7: one complete recommendation service, baseline comparison, evidence and consent checks.
• Weeks 8–9: published pilot release, a selected and approved project domain, restore and rollback checks.
• Weeks 10–12: approved three-college field study and documented fixes.
• Weeks 13–14: working release, guide, aggregate findings and limits, demo video and operational handover.

This is a 14-week template; dates follow the university calendar. Remaining months through month ten cover improvement, expansion and research evaluation. If approvals are delayed, report the field study as incomplete rather than treating synthetic examples as field evidence.

The AI contribution

Start with the existing local embedding model: compare it with keywords, then evaluate improvements using graph evidence. Retrieve useful pathways and explain their sources and limits. Structured project-description extraction or a GraphRAG assistant can follow a bounded evaluation and data-flow review. Shipping the product does not require every proposed technique.

Three colleges and HR review

Proposed exploratory recruitment target: 20 participants per college, subject to protocol approval and feasibility, not a powered superiority sample. Measure task completion, time, errors, explanation comprehension and disclosure comfort; evaluate Arabic/English retrieval on a frozen test set. Two HR experts may review synthetic scenarios and explanations once participation is confirmed; student profiles are not automatically shared.

Publish aggregate results, limitations and the tested release only after running the study. Break down findings by discipline only where group sizes protect participants.

Working files and meetings

Meet weekly for 30–45 minutes and demonstrate an integrated release every two weeks. Assessment covers the problem, implementation, test evidence and reproducible results.

Detailed roadmap · First-two-week task board · Draft evaluation protocol · Validation status

Begin with synthetic data. Health, financial and violation records and personality inference from student accounts are outside scope. Participant research, central storage and external model processing require appropriate review and approval before activation.