4 Downloads Updated 2 weeks ago
Deterministic AI evaluation engine with multimodal extraction, strict reproducibility, and enterprise‑grade governance. Part of the S.L.A.V.K.O.™ Integrity Stack by Formatdisc.
ollama run mladen-gertner/slavkoscore-4.0:deepseek
Models
View all →Readme
S.L.A.V.K.O.™ Score 4.x
Deterministic AI Evaluation Engine for Enterprise Governance
S.L.A.V.K.O.™ Score is a deterministic, auditable, and enterprise‑grade AI evaluation framework designed for high‑integrity environments.
It provides a complete governance pipeline with strict reproducibility, multimodal extraction, and compliance‑ready scoring.
Overview
S.L.A.V.K.O.™ Score is part of the Formatdisc Integrity Stack — a suite of tools engineered for deterministic AI operations.
The system enforces strict reproducibility, schema validation, and cryptographically verifiable audit trails.
Architecture Components
| Component | Purpose | Stage |
|---|---|---|
| SlavkoKernel v8 | Governance & Presentation Layer | Final Output |
| SlavkoShell 2.0 | Validation & Routing Layer | First Checkpoint |
| SlavkoFusion 1.0 | Multimodal Extraction Layer | Second Checkpoint |
| SlavkoScore 4.x | Evaluation & Compliance Layer | Third Checkpoint |
Key Features
Deterministic Execution
- Identical input always produces identical output
- Temperature=0, top_p=0 enforced
- Immutable audit chain with cryptographic signatures
- Full reproducibility for compliance and regulatory audits
Multimodal Support
- Text, image, PDF, UI mock‑ups, and code analysis
- Unified feature extraction pipeline
- OCR and layout analysis
- Vision model support (phi3‑vision, LLaVA, etc.)
Enterprise Security
- Zero‑trust architecture
- Schema validation at every stage
- KernelBus v2 audit trail
- Self‑hosted, no vendor lock‑in
Plugin System
- Custom scoring rules
- Custom extractors for new modalities
- Hot‑reloadable plugin architecture
- Python‑based plugin development
Quick Start
git clone https://github.com//SlavkoScore cd SlavkoScore
pip install -e .
python -m slavko_score.api
Dashboard:
http://localhost:8000
Ollama Integration
Pre‑configured Modelfiles are included for:
- deepseek-r1
- phi3-vision
- qwen2.5:14b
ollama pull deepseek-r1 python -m slavko_score.cli –model deepseek-r1
Documentation
- docs/README_SlavkoKernel_v8.md
- docs/README_SlavkoShell_2_0.md
- docs/README_SlavkoFusion_1_0.md
- docs/README_SlavkoScore_4_x.md
- docs/API_Reference.md
- docs/Determinism_Guide.md
- docs/Deployment_Guide.md
- docs/Plugin_Development_Guide.md
Deployment
Docker
docker-compose up -d
Kubernetes
kubectl apply -f k8s/
Benchmarks
| Model | Latency | RAM | Use Case |
|---|---|---|---|
| deepseek-r1 | 1.5–3s | 8–10GB | High‑precision evaluation |
| qwen2.5:14b | 1–2s | 6–8GB | Standard compliance |
| llama3.1:8b | 0.8–1.5s | 4–5GB | Fast screening |
Security & Compliance
- SOC 2 Type II ready
- GDPR compliant
- HIPAA compatible (with proper configuration)
- Full audit trail for every decision
- Cryptographic signature support
License
MIT License.
See LICENSE for details.
Contact
Formatdisc — Computer Programming & Advanced Software Systems
Founder & System Architect: Mladen Gertner
Web: https://formatdisc.hr
Email: mladen@formatdisc.hr
Phone: +385 91 542 1014
Location: Zagreb, Croatia
OIB: 18915075854
From Code to Governance — One Score, One Audit Trail.