Turn AI into
real returns.

Private AI on infrastructure you control. We train models on your data, deploy them for production use and connect them to your workflows. Hire us for the part you need.

Private by design Built for your stack Support after launch
Example workflows
Select a use case

A model endpoint inside your environment.

Deploy an existing model, measure it on your workload and connect your application.

Inference example
  1. Select the modelOpen weights or your trained model
  2. Deploy the engineOn owned or rented compute
  3. Test the workloadLatency, memory and throughput
  4. Expose the APIAn endpoint your software can call
Data access and network boundaries defined with youAvailable without training or application development
Training  /  Inference  /  AppliedAvailable separately or together

Engineering experience
with teams at

GE21 Geotecnologias OmniHunter ObotX MyArchitectAI North of Zero Kexxu Robotics

Solutions

Choose Training, Inference or Applied. We can work on one part of an existing system or deliver the full workflow. Training is only needed when an existing model does not meet the task.

01Adapt the model

Training

Adapt a model to your data when its existing performance is not enough. We prepare the training pipeline and evaluate the result against an agreed baseline.

  • Pretraining and fine-tuning
  • Vision and multimodal models
  • Evaluation, weights and training code
Discuss training
02Run it privately

Inference

Run open or custom models on compute you own or rent. We deploy the endpoint and optimize it for your workload, then maintain it under an agreed support scope.

  • Private model serving and APIs
  • Latency, throughput and memory optimization
  • Monitoring, updates and support
Discuss inference
03Put it to work

Applied

Connect a model to the work your team does. We build the application, data pipeline and integrations around an existing or newly deployed model.

  • Private document search and processing
  • Agents with defined actions and review steps
  • Vision pipelines and product integrations
Discuss an application

Choose the arrangement that fits the work. Each proposal sets out the scope, responsibilities and costs.

Project delivery

Build a system.

For a workflow, integration or product feature with a defined outcome.

  • A scoped implementation and cost estimate
  • Milestones with review and acceptance criteria
  • Deployment, documentation and handover
  • Code and artifact access specified in the agreement
Discuss the scope
Ongoing engineering

Maintain and improve it.

For a system that needs monitoring, updates or further development.

  • Monitoring and issue resolution within agreed coverage
  • Model evaluation and deployment updates
  • Performance and inference cost reviews
  • Further development planned against your priorities
Discuss ongoing work

Hardware, hosting and external service costs are separate. Support hours and response targets are defined in the agreement.

Selected experience

Two examples of model training and private deployment. Technical measurements are included below.

GE21 · Client engagement · Geospatial

Land-cover classification from satellite imagery.

Need
Classify land use and land cover from panchromatic satellite imagery.
Delivery
A segmentation model trained and evaluated on the client's dataset, running on client-controlled compute.
Technical result

Unet++ achieved 0.92 mean intersection over union on the client's dataset.

Confidential engagement · Inference

Private inference for document reasoning.

Need
Run an open-weight reasoning model over confidential documents in a client-controlled environment.
Delivery
A vLLM deployment on rented GPU compute, measured for a continuous inference workload.
Deployment measurements

Reported throughput: 65 tokens per second on a GPU rented at $0.33 per hour. At sustained throughput, approximately 709,000 tokens per compute dollar. Engineering, idle time and other operating costs are excluded.

Private AI

Private AI applies across Training, Inference and Applied. We define where data is stored, where models run and which systems they can access.

Data access

Agree on approved data sources, storage locations and permissions before development.

Compute

Owned servers, rented GPUs or on-device deployment where the workload fits. Infrastructure costs stay visible.

System boundaries

Define API access, network requirements and external dependencies. Offline operation can be assessed during scoping.

The company

Cicotoste Technologies is an AI engineering firm based in Brazil, focused on private AI workflows. We offer Training, Inference and Applied as separate services or as a combined engagement.

Gabriel Cicotoste founded the firm and leads the technical work. His experience includes geospatial models, production inference and 3D perception.

Meet the founder

FAQ

Do we need all three services?

No. You can hire us for Training, Inference or Applied on its own. An existing model can go straight to deployment, and an application can use an endpoint you already run.

Do we need to know which model to use?

No. Tell us what the system needs to do. We assess model options using your data and requirements.

Can you work with what we already have?

Yes. We can improve an existing model, optimize its deployment or add an integration.

Can the system run on our infrastructure?

Yes. We deploy on owned or rented servers. Data access and any external services are agreed during scoping.

Is this a project or a monthly service?

The initial build is a scoped project. Ongoing maintenance and development can be covered by a monthly agreement.

How do you measure the return?

We compare the system with the current process. Depending on the project, we track time per task, operating cost or output quality. A financial return calculation also needs your business costs and usage data.

What do we own at handover?

The agreement specifies what code, documentation and model files you receive. Third-party software and models retain their original licenses.

Let's talk

Discuss your project.

Walk us through the process you want to automate or the system you want to build. We will discuss the data, constraints and likely scope.

Book an introductory call 30-minute introductory call.