Atomic AoL logo Atomic AoL Accreditation & AssuranceOfLearning · a project by Marino&Carli
How it works

One operator. One wizard.
Forty-one specialists.

Atomic AoL replaces weeks of manual compilation with a guided web wizard and an agent pipeline that executes the whole AACSB Standard 5 workflow — under human control at every decision that matters.

Step by step

The operator's journey

1

Set up your tenant

Your institution gets an isolated workspace. Every row of data is stamped with your tenant ID server-side — agents cannot read or write across institutions.

2

Seed the standard

An agent ingests the AACSB standard text and indexes it semantically, so every later judgement can cite the standard it serves.

3

Connect your curriculum

Agents read your public programme pages and module descriptions, extract qualifications and learning outcomes, and propose the mapping from competency goals to modules.

4

Upload your evidence

Grades and indirect measures (alumni surveys, exit surveys, employer feedback) arrive as simple CSV files — templates included, validation messages in plain language.

5

Approve the judgement calls

Proposed competency goals wait in an approval queue. Nothing proceeds until a named human approves — and the approval is recorded with name and timestamp.

6

Run the pipeline

One click. Layer 2 computes cohort aggregates, multi-year trends and benchmark breaches; Layer 3 writes the narrative and signs the bundle. Minutes, not weeks.

Under the hood

Three layers, each with one job

Layer 1 · Connectors

13 agents

Scraper, AACSB seeder, competency-goal generator, modules reader, outcome mappers, coherence gate. They build the curriculum graph the analysis stands on.

Layer 2 · Analytics

19 agents

Grade collector, cohort aggregator, trend computer, benchmark comparator, narrative-arc balancer and colleagues. Deterministic maths where maths belongs; AI judgement only where judgement is needed.

Layer 3 · Narrative

9 agents

Executive summariser, evidence narrator, gap flagger, report stitcher, accreditation packager. Output: a typeset PDF plus the evidence trail that backs every sentence.

What you provideWhat the pipeline produces
Programme / module URLs (public pages)Qualification & learning-outcome graph
Grades CSV (per cohort, per year)Cohort aggregates + multi-year trends per competency goal
Indirect measures CSV (surveys)Indirect vs direct alignment analysis
Benchmark policy (or let the AI propose one)Breach analysis with severity grading
~30 minutes of your attentionSigned submission bundle: REPORT.PDF + EVIDENCE.JSONL + SHA-256 MANIFEST
Human in the loop

The AI proposes. Your faculty decides.

Atomic AoL is built around approval gates, not around autonomy. Competency goals — the intellectual anchor of the whole report — are blocked until someone with a name approves them. The same pattern guards key analytic judgements. This is what makes the output defensible in front of a peer-review team: every claim has a human who said yes, and a timestamp that proves when.

Approval queue Named sign-off Timestamped decisions Full audit trail

Watch it run on a synthetic school

We demo the full pipeline on "Hogwarts Business School" — six years of synthetic data, so you see real mechanics with zero privacy concerns.

Book a pilot → Book a call →