18 repositories scanned. 5 AI systems discovered across code and ML services. This is a sample — sign in to run Spekris on your own organization.
acme/claims-engine · .env.example
Confidence
94%
Production ML model scoring every claim submission for fraud. Confidence threshold 0.85.
Finding: No documented owner or model card. If this model drifts or produces biased outputs, claims are being approved or rejected with no human review and no accountability.
acme/ml-services · ml/nlu-triage/requirements.txt
Confidence
96%
Fine-tuned Bio_ClinicalBERT model classifying urgency of patient messages. Runs as a FastAPI service.
Finding: Owner @achen has not committed in 203 days and may have left the company. A patient triage model with no active owner is a governance and patient safety risk.
acme/platform · experiments/prototype-chat/chat-server.ts
Confidence
97%
Patient-provider WebSocket chat prototype. No authentication. In-memory only. No audit logging.
Finding: If this prototype ever processed real patient data in this state, it may create significant HIPAA exposure. Original author inactive for 189 days.
acme/platform · experiments/feature-flags/flag-config.json
Confidence
88%
AI-powered ICD-10 and CPT billing code suggestions. Active at 20% rollout. Fine-tuned model.
Finding: No documented accuracy baseline. No human review step before codes are used. Inaccurate suggestions could create billing compliance risk.
acme/claims-engine · cron/weekly/run-anomaly-detection.ts
Confidence
91%
Weekly ML-based anomaly detection on all claims. Checks upcoding, duplicate billing, unusual volumes.
Finding: Owner is documented only in a code comment. If James Liu has left, there is no accountable owner for a system that makes compliance decisions every Sunday.
Connect your own sources and Spekris will run the same scan on your real code, cloud, and automations. First scan takes under five minutes. No agent to install.
No credit card required. Credentials encrypted with AES-256-GCM.