Find what your work
actually needs.
The Assessment Engine compares models on your work and recommends where to run them. It checks cost, quality, data requirements and the hardware each option needs.
From source data
to a decision.
Collect evidence
The Engine reads API and gateway logs, provider usage, billing and token data, alongside prompts, workflows and infrastructure information. It establishes which models handle the work and what they cost.
Prepare the data
The Engine anonymises sensitive records, clean and deduplicate the data, and group similar workloads. It maps the privacy constraints and quality requirements before comparing the options.
Compare the options
The Engine benchmarks current, cloud and open-weight models on the same tasks. It blind-tests their responses and compare API costs with the infrastructure needed to run private AI inference.
Recommend a setup
The Engine produces model and routing choices, a costed deployment plan and an environmental projection. It recommends keeping the existing setup where a change does not improve the result.
Insights beyond price per token.
View comparison
| Requirement | What is compared | What the recommendation answers |
|---|---|---|
| Cost and usage | Provider bills, token use, workload volumes and infrastructure costs. | Where to reduce spend within the current provider, route between models or run suitable work privately. |
| Output quality | Current and candidate responses on the same work, tested blind against the task’s agreed threshold. | Which alternatives meet the requirement, rather than simply costing less. |
| Infrastructure fit | The workload mix and volume against the model and inference capacity needed to serve it. | Whether existing capacity still fits, a different model mix is needed, or private inference is justified. |
| Data and impact | Approved data routes, compute requirements, energy coverage and grid carbon intensity. | Where the work may run and the estimated operational impact of each suitable option. |
A plan you can
act on.
The output connects the evidence to a specific model, routing and deployment recommendation. Savings and environmental projections remain estimates until checked in production.
- Savings, environmental projections and an optimisation or deployment plan.
- Model mix, routing architecture and inference-capacity recommendations
- Benchmark results, blind test comparisons, and all the data and evidence collected