OCMD AI Explorers Summer Camp - Grades 9-12

OC Music & Dance (OCMD), in partnership with Pacific Solutions

501(c)(3) community arts and technology education center

Irvine, Orange County, Southern California, USA

Lead Master Coach supported by trained Assistant Coaches under a two-deep adult staffing model

• On-site nonprofit community program
• Summer day camp

Seven-day intensive; afternoon cohort from 1:00 PM to 4:00 PM

August 1-8, 2026; approximately 19.5 contact hours

18 enrolled students

One program-provided Mac laptop per learner on a password-protected network

• No individual student vendor accounts
• All prompt runs required coach review
• Afternoon scheduling competed with work, athletics, and transportation
• Advanced concepts had to fit three-hour blocks

Primarily Grades 9-12, approximately ages 14-18; the source notes that a small number of younger learners joined the cohort

No prior AI experience required

No programming prerequisite reported

6 of 18 had perfect attendance; 9 of 18 attended at least five days and the Day 7 showcase; 9 matched pre/post survey pairs

• Direct advanced AI architectures with technical rigor
• Build multi-layered system prompts and multi-scenario assistants
• Evaluate model bias, context limits, and synthetic-media authenticity
• Apply academic-integrity and disclosure practices

• Design agent decision trees with human checkpoints
• Vibe code interactive web tools
• Assemble portfolio-grade capstones with formal disclosure documentation

• Not unsupervised access to commercial AI services
• Not a controlled efficacy trial
• Not a professional software-engineering credential

Coach-operated text, image, video, automation, assistant-building, and AI-assisted coding tools

• Claude
• ChatGPT
• Gemini and Veo
• Automation blueprints
• HTML and JavaScript vibe-coding templates

• Generative partner
• Object of critical evaluation
• Support for assistant, agent, and prototype design

• Students composed prompts and plans in a de-identified workspace
• Coaches reviewed prompts with a Sideline Check before execution
• Coaches ran approved prompts and returned outputs for student iteration

• Coach Proxy Model
• Fictional Character Default
• Personal-data redaction
• Parent or guardian acceptable-use consent
• Human review before output display
• Scheduled 180-day data purge

• Day 1 foundations and orientation
• Day 2 advanced prompting
• Day 3 image generation and art direction
• Day 4 video and multi-scenario assistants
• Day 5 complex agent design and AI-assisted web building
• Day 6 ethics, trust, and critical thinking
• Day 7 academic-integrity capstone and showcase

• Students set goals, composed prompts, evaluated outputs, and documented human authorship decisions
• Coaches approved and executed prompts, checked outputs, and documented incidents
• AI systems generated drafts and media within human-defined constraints

• Video-debrief-practice rhythm
• Worked examples and pre-filled templates
• Four-step Compose, Review and Run, Return, Iterate loop
• Advanced 4 Rs prompt cards and formal disclosure requirements

• Half of enrolled learners did not complete five of seven days
• Summer employment, athletics, and transportation contributed to attrition
• Some students resisted post-showcase surveys
• Complex concepts required careful pacing
• Day 1 late arrivals compressed the orientation
• The cohort raised questions about responding to disengaged learners

• Split prompt approval and execution duties across coaches
• Added worked examples and structured prompt templates
• Disabled unnecessary communication features
• Emphasized formal AI-use disclosures
• Corrected lesson sequencing on the course website
• Recommended a longer orientation, better audiovisual support, reference images, and quick student-work showcases

• 9 of 18 learners attended at least five of seven days and the Day 7 showcase
• All nine showcase participants completed post-program evaluations
• Every final capstone included an AAB Standard Disclosure Statement

• Average accuracy across objective AI scenarios rose from 73% to 98% among matched participants
• Four of five post-program scenarios reached 100% correct; the fifth reached 88%
• Self-efficacy for writing effective prompts rose from 3.25 to 4.35
• Self-efficacy for improving prompts strategically rose from 3.52 to 4.35
• Viewing prompting as a creative process rose from 2.25 to 3.63

• Matched results represent nine retained learners, a highly engaged subgroup
• Results are descriptive and not based on a randomized comparison
• Attendance and evaluation attrition limit generalization to all enrollees

• Learners used de-identified seat codes rather than names
• No learner held a personal commercial AI account
• Real-person likeness or voice generation required separate signed consent
• Raw creative outputs, media, and school records remain withheld from public release
• Student workspaces are scheduled for deletion 180 days after camp

• Two-key student pledge and parent consent
• Fictional Default guardrail
• Human-in-the-loop review
• Incident and hallucination logging
• Required disclosure statement for final presentations

• De-identified student workspace
• Private coach compliance ledger
• Observer and coach feedback
• UCI interim matched pre/post evaluation dataset
• Daily physical attendance records
• Capstone showcase presentations

• Coach-mediated generative AI instruction for high-school learners
• Age-band differences in attendance and survey retention
• Academic-integrity disclosure protocols
• Operational use of the 4D AI Fluency Framework
• Human-review throughput in multi-learner classrooms

• AI literacy
• Learning sciences
• Educational technology
• Academic integrity
• Human-in-the-loop AI governance

• Completed
• Approved for public registry publication
• Aggregate and de-identified disclosure only

• AAB-PILOT-2026-AI_Explorer_Program-002
• AAB-PILOT-2026-OCMD-001

Grades 9-12

Community-based summer camp

Generative AI, multimodal generation, assistants, agents, and AI-assisted coding

Coach-mediated guided practice and project-based capstone

Moderate with documented controls

Low to moderate; de-identified learner work and evaluation data