Argus is an 8-billion parameter LLM engineered from the ground up for Oracle, Workday, and enterprise applications. It doesn't just automate—it reasons, troubleshoots, and orchestrates with deep domain expertise.
Through domain-adaptive training and curriculum learning, Argus achieves state-of-the-art performance in ERP tasks—outperforming GPT-4.1 while being 29× more efficient than comparable large models.
Built for Enterprise Complexity
Unlike general-purpose LLMs, Argus understands ERP-specific concepts, workflows, and the multi-step reasoning required for real-world operations.
Domain-Specific Knowledge
Trained on 1.5B tokens of ERP-specific content including Oracle documentation, API references, database schemas, and expert knowledge.
1.5B
Training Tokens
Multi-Step Reasoning
Advanced Chain-of-Thought reasoning adapted for ERP contexts—analyzing workflows, troubleshooting errors, and orchestrating cross-module processes.
81.4%
Benchmark Accuracy
Agentic Capabilities
Autonomous monitoring, exception handling, and workflow orchestration. Proactively flags issues and triggers downstream processes.
24/7
Autonomous Operation
Conversational Intelligence
Natural instruction-following and multi-turn conversations. Understands context and provides structured, actionable guidance.
92.3%
Relevance Score
Training Methodology
Three-Pillar Training Recipe
Our sequential training strategy combines domain-adaptive pre-training, curriculum learning, and preference optimization.
Phase 1
Domain-Adaptive Pre-Training
Exposing the model to ERP-specific concepts while preserving broad language capabilities through strategic data mixing.
1.5B tokens of curated ERP content
70:30 domain-to-general ratio
Extended 40K token context
Multi-GPU clustering pipeline
Phase 2
Multi-Phase Supervised Fine-Tuning
Progressive curriculum learning from basic instruction-following to advanced reasoning and tool integration.
230K instruction-response pairs
Curriculum learning approach
Multi-turn conversation synthesis
Cross-module integration scenarios
Phase 3
Reasoning Enhancement
Advanced preference optimization using expert-curated Chain-of-Thought traces for complex analytical tasks.
17K Oracle reasoning traces
Direct Preference Optimization
Self-correction capabilities
Adaptive reasoning patterns
Benchmark Results
Validated on 10,000 ERP Tasks
Comprehensive evaluation across finance, supply chain, HR, and cross-module integration scenarios.
81.4%
Overall Accuracy
Matches 235B parameter models
92.3%
Domain Relevance
Highest domain-specific score
81.2%
Completeness
+11.5% vs baseline
82.5%
Helpfulness
Production-ready utility
Model Comparison (ERP Benchmark)
Argus-8B
8B
Ours
81.4%
GPT-4.1
Large
78.4%
Qwen3-235B
235B
81.4%
Qwen3-8B (Base)
8B
75.9%
*Evaluated on 10,000 ERP tasks across diverse functional domains
Domain Expertise
Deep Knowledge Across ERP Modules
Trained on Oracle Fusion, Workday, and enterprise documentation covering all major functional areas.
