Portfolio Build
Dyniq
Multi-agent LLM platform — cost-optimal routing across a large model portfolio with automated quality regression detection.
Metrics
- 107
- agents mapped
- 9
- LangGraph StateGraphs
- 5
- model providers
Problem
Route many specialised agents reliably and cheaply across a large model portfolio.
Stack
- 9 LangGraph state machines
- PydanticAI
- RAG over pgvector
- 7-tier routing cascade across 5 model providers (OpenAI, Anthropic, Google, DeepSeek, Moonshot)
- 107 agents mapped to cost-optimal models
- LLM-as-judge (1–10 rubric, 15% regression alarm)
- Langfuse observability
- LiveKit voice
Constraints
Multi-provider cost control without sacrificing quality; regression detection that triggers before it reaches users.
Outcome
Cost-optimal multi-provider routing with automated quality regression detection.