The job
Optimus is our on-prem appliance. It ships preloaded with open-weight models and the full Lyzr agent platform, and it runs a customer's entire agentic system inside their own four walls with zero data egress. Mini is 8x RTX PRO 6000 Blackwell, 768 GB of GPU memory, dual EPYC 9005, 5 to 6 kW in 4U to 5U.
The design is in place and the vendors are picked. We need someone to own the architecture from here: design it properly, test it, and stand behind it in front of customers.
What you will do
- Own the box architecture: GPU, CPU, storage, storage memory, network cards, cooling, power envelope
- Build it, test it, and be honest about what breaks under sustained inference load rather than on a spec sheet
- Be the technical point of contact for customers on networking and cards. When a bank's infrastructure team asks how this lands in their rack, you answer
- Hold the line on silicon neutrality. We are evaluating NVIDIA, AMD and Intel on purpose and are not locking into one vendor
- Work the vendor relationships and the RFQ specs across the tier line
- Make Optimus Mobile physically real: a cluster that survives a truck, plugs into a customer site, and works
What we are looking for
- You have architected and shipped real systems, not designed them on paper. Servers, appliances, GPU clusters, or dense compute in a constrained envelope
- Deep on power and thermal. 6 kW in 4U is not a rounding error and you know it
- Fluent on high-speed networking and interconnect, and on what enterprise data centers will and will not accept in a rack
- You can hold your own in a customer room
- Comfortable with GPU vendor economics, supply timelines and the gap between announced availability and actual availability
Nice to have
- Built for regulated or air-gapped environments
- Manufacturing or contract manufacturing experience
- Some feel for the inference stack above the metal, since the box exists to serve models