GAISSA

Make your AI
faster and
cheaper.

GAISSA Optimizer selects and applies improvements to your model and how it runs. Our team runs it for you, delivering an optimized model or setup with results measured against your goals.

Let’s talk about your AI

Phi-3.5 Mini

A smaller model, same quality.
More requests at once.

Before

7.14 GiB

With Optimizer

3.80 GiB

Memory used by the model

The optimized model kept its accuracy while using less memory. Changes to how it runs let all eight requests run together.

The same eight requests

4 running

4 waiting

8 running

None waiting

More from
the same AI.

Built on our AI efficiency research at the Universitat Politècnica de Catalunya.

See all research results

Language models · Phi-3.5 Mini

More answers for your compute budget.

60%lower estimated GPU cost
per response.

2.5× as many requests completed per second.
Quality preserved.

How the cost compares

At the same GPU hourly price, the measured speed gain brings the compute cost of the same work from 100% to 40%.

Before
100%
Optimized
40%

This estimates compute cost at the tested load. Your bill also depends on usage and hosting commitments.

Speech recognition · NVIDIA Parakeet

Process more audio on the same GPU.

2.4×as much audio
processed per second.

With transcription accuracy preserved.

Working with GAISSA

  1. Review your setup

    We look at your model, how you use it and what you want to improve. Together, we define the requirements the changes must meet.

  2. Measure and improve

    Our team measures the starting point, applies supported changes with Optimizer and checks the results on representative tasks.

  3. Receive the changes

    You receive the optimized model or system changes, before-and-after results and an explanation of what changed.

What would you like your AI to do better?

Tell us which model you use and what you want to improve, whether that’s cost, speed, memory use or something else.

Let’s talk about your AI
Can you optimize the AI we already use?

GAISSA Optimizer selects, applies and tests improvements to your model and how it runs. We confirm compatibility with your setup before starting. If you only use a third-party model API, we first check whether the available access allows us to help.

Will we need to change models or infrastructure?

Not necessarily. Optimizer can improve the model you already use or the way it runs. Any proposed switch of model, provider or hardware is agreed with you before implementation.

What will we receive?

An optimized model, changes to your system, or both, with before-and-after measurements and an explanation of what changed. Our team operates GAISSA Optimizer for you.

Will optimization affect quality?

You set the quality requirements and any tradeoffs you are willing to accept. Optimizer tests changes against those requirements by comparing the original and optimized setup on representative tasks. Changes that fail are rejected.

What do you need from us?

Your model, how it runs and what you want to improve. To apply and measure changes, our team also needs access to the relevant model or system and examples of the tasks it performs.