Model orchestration
Aucert uses a tiered model strategy to balance quality and cost. Different pipeline stages use different model capabilities.
Model tiers
| Tier | Description | Cost/1M tokens | Use case |
|---|---|---|---|
| Tier S | Self-hosted fine-tuned | $0.01-0.05 | High-volume, domain-specific tasks |
| Tier API | Claude Sonnet (API) | ~$3.00 | Complex reasoning, visual analysis |
MVP strategy
Phase 1 uses a single model (Claude Sonnet via API) for all pipeline stages. No routing logic — all requests go through the same model.
Planned routing (Phase 2+)
A RouterPolicyEngine will dynamically route requests based on:
- Task complexity (simple classification vs multi-step reasoning)
- Required capabilities (text-only vs multimodal)
- Cost budget per customer
- Latency requirements
LLM API access
Current configuration:
- Primary: AWS Bedrock (pending Activate credits)
- Fallback: Anthropic Direct API
- Blocked: Azure AI Foundry (not available on Founders Hub credits)
What's next
- 5-layer deep dive — Full pipeline architecture
- Verification Cascade — Multi-stage verification