DR. SAEID HOMAYOUN / Accounting · Auditing · AI Research
DARWIN–AuditBERT · Academic Research Laboratory

Explainable analysis of Critical
and Key Audit Matters.

Research demonstration for CAM/KAM disclosure screening, accounting-topic mapping, and transparent evidence identification. Developed under Dr. Saeid Homayoun, University of Gävle, Sweden. This application processes text locally in your browser and makes no external inference calls.

CAM · PCAOB / US context KAM · IFRS context No credentials or private data required Demo only · NOT a trained AI model
01 / INPUT

Screen an audit matter

Use the synthetic examples or an approved public audit-report excerpt. Do not paste confidential client records into public research demonstrations.

Method: a fixed, auditable keyword dictionary. Neither a FinBERT classifier nor a Llama model is invoked.

02 / RESULTS

Accounting-topic evidence

Results show exact vocabulary matches and an illustrative accounting reference; the hit count is not a probability, risk score, or audit-quality assessment.

TopicHitsTerms matchedReference
Analyze a matter to view vocabulary matches.
Awaiting input. Every resulting classification requires source verification and expert interpretation.

FinBERT · Topic classifier Not trained

Independent domain adaptation and issuer-disjoint evaluation are proposed. No model weights or new accuracy scores have been published under SADHON.

Llama 3.1 8B · Responses Not trained

Separate QLoRA response-generation research requires licensed examples, GPU training, human validation and explicit safeguards against fabricated procedures.

Reproducible research Code staged

Existing notebooks and raw datasets remain in private Google Drive. Public-safe methods and a new TF–IDF benchmark are under GitHub version control.