LARS Training Curriculum
Knowledge Domains to Train
| Domain | Source | Priority | Status |
|---|---|---|---|
| Identity | Who LARS is, Corlera, Chris Foust | High | ✅ Done |
| Task Reasoning | Multi-step operations, ambiguity | High | ✅ Done |
| Nexus Operations | Workflow, tools, environments | High | Next |
| Security | Cybersecurity, monitoring, protection | High | Planned |
| Training Knowledge | 3D format, how to train models | High | Planned |
| Client Customization | Adapting training for clients | Medium | Future |
| Internet Search | Web access, research capability | Medium | Future |
Workflow → Training Pipeline
Critical Insight: Everything in Workflow should become LARS training data.
- Voice protocols → LARS knows how to communicate
- Tool schemas → LARS knows what tools exist
- Environment info → LARS knows the infrastructure
- Work patterns → LARS knows how we operate
- Credentials patterns → LARS knows security (Locker, not plain text)
Action Item: When updating Workflow, flag items for LARS training.
Security Training Topics
- Network Monitoring
- Traffic patterns, anomaly detection
-
What's normal vs suspicious
-
Access Control
- Who should access what
-
Authentication awareness
-
Data Protection
- What's sensitive, what's not
-
Credential handling (Locker only)
-
Incident Response
- What to do when something's wrong
- Alert patterns, escalation
Training Knowledge (Meta-Level)
LARS needs to understand: 1. What 3D format is and why it works 2. How to structure training examples 3. Optimal parameters (epochs, LR, batch size) 4. How to evaluate training success 5. How to create client-specific datasets
This enables LARS to train other models autonomously.