A two-week proof of concept engagement that answers the model and hardware questions before you commit to on-prem infrastructure. You evaluate candidate models against your actual work. We handle the curation, the environment, and the benchmarks.
If you're considering on-prem AI for your team, you're navigating two questions with not much data to go on. Will an open-weight model meet your quality standards on your actual work? And what hardware does that model need to run well for a team your size?
These are hard questions to answer without real equipment and real testing. Most teams guess, overbuy, or do nothing. This PoC exists because these questions are difficult to answer.
Before the PoC starts, we narrow the universe of open-weight models down to 2-4 candidates that are realistic for your use case, your budget, and the hardware footprint that would make sense for a team your size. We then stand up a chat interface and/or coding assistant, point it at those candidates, and hand it to your team for two weeks to try against their actual work. You decide whether any of them clear your bar.
We take the top candidate from Phase A, run it on dedicated GPUs that matches the hardware profile we'd recommend, and measure throughput, latency, and concurrent capacity under realistic load. This is where we answer what hardware is actually required to run the model for your team.
Everything produced during the PoC is yours, whether you proceed to deployment or not.
An evaluation environment your team can use for two weeks. Chat interface and coding assistant, pointed at candidate models we've pre-selected for your use case and budget.
A written report covering the candidate models, measured performance benchmarks, hardware recommendations at three price points, and a deployment plan.
A recommendation call to walk through findings and your team's feedback.
A formal install proposal if you decide to proceed.
Every PoC ends in one of four outcomes, based on your team's evaluation and the benchmark data.
Your team finds a candidate model that meets their bar. We have the hardware spec and benchmarks ready.
The models work for some of what you want to do but not all of it. Your team and ours map out which parts of the workflow are a fit and which aren't.
None of the candidates quite clear your bar, but we're seeing model releases on a monthly cadence and the trajectory suggests revisiting in 3-6 months. We tell you what to watch for.
Your use case needs capabilities only the frontier closed models offer today. Better to know now than after buying hardware.
You have a specific use case for AI, whether that's coding assistance, document processing, an internal knowledge base, or something similar.
Your customers, compliance posture, or contracts make sending code or data to third-party AI services complicated.
You're seriously considering on-prem or private-cloud AI, not just exploring whether AI is worth using.
You're unsure whether current open-weight models are good enough for what you need.
Tell us what you're working on and we'll send a scoped proposal with pricing, timeline, and specifics.
Request a proposal