We're working on a problem that sits at the intersection of distributed systems and ML deployment: how do you get multiple AI agents — backed by different models, running across different machines — to coordinate work, verify each other's output, and produce results you can actually trust?
OmniNode is our answer. The architecture is:
- Contract-driven task delegation (structured specs, not prompt chains)
- Multi-model routing — local inference (Qwen3-Coder-30B on RTX 5090, DeepSeek-R1 on M2 Ultra) alongside cloud models, routed by task characteristics
- Deterministic execution with event sourcing (Kafka) and replayable state
- Evidence-backed completion — work isn't "done" because an agent said so, it's done when durable verification passes
The core insight we're building on: as agent autonomy increases, reproducibility and enforceable correctness matter more, not less. Prompts don't scale. Contracts do.
Looking for engineers who think in systems, care about how things fail under real conditions, and are comfortable working across the ML/infrastructure boundary.
Fully remote. DM me with something you've built or something interesting you've been experimenting with.
[link] [comments]