NSCC Curriculum Outcome Mapping & Audit Pipeline — Full Institutional Sweep
The NSCC Curriculum Outcome Mapping & Audit Pipeline completed a full institutional sweep of the college's academic inventory, evaluating every course learning outcome across every accredited program — entirely on-premises using locally hosted open-weight AI inference. The pipeline processed 15,459 Learning Outcomes across 3,225 courses and 148 academic programs, producing publication-ready accreditation compliance reports for each, at a direct cloud API cost of $0.00.
This report documents pipeline accomplishments, infrastructure resilience under adverse conditions, the full token workload and equivalent commercial cost, and the strategic value of a local AI architecture for ongoing curriculum quality assurance.
alignment.json files for every program.Standards_Compliance_Report.docx) containing compliance breakdowns, cognitive rigor scores, and actionable recommendations for academic chairs and curriculum committees.audit_compliance.py against open-weights Qwen2.5-32B-Instruct-AWQ via local vLLM inference.The pipeline operated under genuinely adverse infrastructure conditions. The following incidents were encountered and recovered from without data loss:
WinError 10065) and caused momentary hypervisor OOM freezes during restart boot loops, temporarily taking the vLLM endpoint offline.run_compliance_batch.py via compliance_audit.json ensured zero-loss recovery. Upon hardware restoration, the batch seamlessly skipped previously audited programs and resumed exactly where processing had stopped — no duplicate processing, no re-evaluation cost.audit_compliance.py caught momentary APITimeoutError events (e.g., on Truck and Transport Repair) and skipped cleanly without crashing multi-hour batch operations.Across both pipeline phases, 8,852,727 tokens were processed entirely on-premises over 27.75 hours of GPU compute on the NSCC Truro Campus AI Cluster (10.30.0.21:8001). No curriculum data left the NSCC private network.
| Workflow Phase | Input Tokens | Output Tokens | Total Tokens |
|---|---|---|---|
Phase 1 — LO/PO Alignment (alignment.json) |
2,391,904 | 734,986 | 3,126,890 |
Phase 2 — Cognitive Compliance Audit (compliance_audit.json) |
3,732,146 | 1,993,690 | 5,725,836 |
| Total Dataset Volume | 6,124,050 | 2,728,677 | 8,852,727 |
The table below shows what this exact 8.85 million token workload would have cost at current 2026 public API pricing from major cloud providers. All API prices are billed in USD; CAD equivalents use an approximate rate of 1 USD = 1.38 CAD. Anthropic pricing sourced from the Anthropic API (2026-06-24); OpenAI pricing approximate as of mid-2025.
| Provider / Model | Pricing per 1M (USD) | Total (USD) | Total (CAD ~) | Actual Cost |
|---|---|---|---|---|
| Anthropic Claude Fable 5 | $10.00 in / $50.00 out | $197.67 | $272.78 | $0.00 |
| Anthropic Claude Opus 4.8 | $5.00 in / $25.00 out | $98.84 | $136.40 | $0.00 |
| Anthropic Claude Sonnet 5 (intro, through Aug 2026) | $2.00 in / $10.00 out | $39.54 | $54.56 | $0.00 |
| Anthropic Claude Haiku 4.5 | $1.00 in / $5.00 out | $19.76 | $27.27 | $0.00 |
| OpenAI GPT-4.1 | $2.00 in / $8.00 out | $34.08 | $47.03 | $0.00 |
| OpenAI GPT-4o | $2.50 in / $10.00 out | $42.60 | $58.79 | $0.00 |
| OpenAI GPT-4o-mini | $0.15 in / $0.60 out | $2.56 | $3.53 | $0.00 |
| Local vLLM — NSCC AIProx Cluster | — | $0.00 | $0.00 | $0.00 |
CAD estimates are approximate. Exchange rate (USD/CAD ≈ 1.38) is indicative and will vary.
Avoiding up to ~$198 USD (~$273 CAD) in direct API fees for this single full-institutional sweep is a concrete near-term saving (equivalent to running the same workload against Anthropic's flagship Claude Fable 5 at current 2026 pricing). The deeper value, however, is structural: local AI inference eliminates per-token cost entirely, turning iterative re-auditing from a budgeted expense into a zero-marginal-cost operation.
Internal curriculum documents, draft pedagogical outcomes, and program evaluations remained entirely within the NSCC private network (10.30.0.x). No institutional IP transmitted to any third party.
As academic chairs refine outcomes based on compliance report recommendations, re-running the full pipeline costs nothing in API fees. Iteration is free.
The idempotent batch architecture supports scheduled runs on a nightly or per-semester basis — enabling Continuous Integration checks for curriculum quality at no additional cost per run.
Adding a new program requires only dropping source documents into the data/raw/ folder. The pipeline generalizes automatically — no code changes, no additional cloud budget.
Standards_Compliance_Report.docx contains a prioritized list with specific recommendations.