Physical AI / Computer vision / Edge and embedded systems
Perception systems that run in production, from edge to cloud.
SentiaLab designs, trains and deploys computer vision and AI systems, from annotated data to optimized software running on edge devices, on-premises servers or in the cloud.
A typical perception pipeline, from sensor to decision
- Cameras and sensors
- Ingest and decode
- Inference
- Tracking and rules
- Events and APIs
Deployed on edge devices, on-premises servers or in the cloud.
Selected work
Client work
Transportation
Wheel counting for free-flow tolling
A free-flow tolling system needed to count vehicle wheels from highway RTSP cameras in real time. We built the annotation workflow, trained the detection models and wrote the C++20 pipeline, deployed on Jetson Orin NX for production use on highways.
C++20 · GStreamer · DeepStream · YOLO · TensorRT
Healthcare
Action detection for elderly care
Software for elderly care needed to detect actions and events to help nurses and caretakers. We trained and tested the neural networks, ported the software to Jetson devices and integrated C++ and Rust for performance-critical paths.
PyTorch · CUDA · C++ · Rust · Yocto
Research
Vehicle classification with CNNs
Research on classifying vehicle categories from optical-curtain profiles with convolutional neural networks, co-authored and published in IEEE Access (2024).
Read the paperPyTorch · OpenCV
Earlier engineering experience
Delivered by our founder before SentiaLab.
Automotive
Connected vehicle control over CAN J1939
An IoT system in which several in-vehicle controllers communicate over CAN J1939 to operate safely and send data to the cloud. C++ applications for embedded Linux, Yocto recipes, and the packaging, build and unit-test setup.
C++ · Yocto · CAN J1939 · CMake · Google Test
Semiconductor
Sensor software for semiconductor equipment
Software for a sensor subsystem in semiconductor manufacturing machines: hardware integration, integration across subsystems, the internal testing framework and the CI/CD pipelines.
C · C++ · Python · GitHub Actions
Industrial IoT
Machine-monitoring gateway for OEE
A gateway that collects data from industrial machinery for Industry 4.0 integration, with software to analyse overall equipment effectiveness. Electronics, firmware, and the data collection and analysis software.
C/C++ · ESP32 · Python · C#
- Sectors
- Transportation, healthcare, industry
- Deployment
- Edge, on-premises, cloud
- Delivery
- Remote, worldwide
- Based in
- São Paulo, Brazil
Services
Edge AI
We design, train and deploy optimized neural networks for real-time inference, on constrained edge hardware or wherever the workload runs best.
- Custom model training for your use case
- Model export and optimization for the target runtime
- Deployment on embedded GPUs and edge devices
- Real-time video analytics pipelines
Computer vision
Full vision pipelines, from data collection and annotation to training and deployment.
- Object detection and classification
- Action recognition
- Annotation pipeline setup and management
- Production-grade vision applications
Embedded systems
Reliable software for constrained platforms: embedded Linux applications, firmware and hardware integration.
- Embedded Linux applications and custom images
- Firmware development
- Industrial and vehicle bus integration
- Cross-compilation and packaging
MLOps and deployment
The infrastructure that takes a model from a notebook to production, on a device, on-premises or in the cloud.
- CI/CD pipelines for models and software
- Containerized build and runtime environments
- Automated testing and validation
- Cluster orchestration for development and production
How we work
Step 1
Assess
We review the problem, the data you have and where the system must run, and agree on what success means.
Step 2
Prototype
We train a first model and run it on the target platform early, so feasibility is proven in the real environment.
Step 3
Optimize
We tune the model and the pipeline for the target platform until accuracy, latency and cost meet the targets.
Step 4
Deploy
We deliver production software with CI/CD and tests, and hand it over to your team.
Ways to work with us
Fixed-scope project
A defined system, designed, built and delivered end to end.
Audit and optimization
A review of an existing model or pipeline, with the changes needed to make it faster, cheaper to run or deployable on the target platform.
Embedded engineering
An engineer working inside your team for a period, on your roadmap.
Who we work with
Product companies
Adding a vision or AI feature to a robot, a machine or a software product.
Infrastructure and industrial operators
Systems that have to run in the field: tolling, traffic, factories.
Startups
Teams without ML engineers that need the whole pipeline built.
Systems integrators
Partners who need the vision or embedded part of a larger delivery.
Company
Sentia, from the Latin sentīre: to feel, to perceive.
SentiaLab is an engineering company based in São Paulo, Brazil, working remotely with clients in Brazil and Europe. We build systems that perceive the world and run wherever the job demands: on the device, on-premises or in the cloud.
We cover the whole path: annotating data, training models, optimizing them for the target platform and shipping the software that runs in production. We design for the deployment constraints from the first day, rather than adapting a model afterwards.
Founder
Caio Piccirillo
Founder and principal engineer
- 9+ years in embedded systems, 6+ in computer vision and edge AI
- Co-author of research published in IEEE Access (2024)
- Former university lecturer in AI
- Works in English, Portuguese, Spanish, Dutch and French
Have a perception problem that has to work in the real world?
Tell us what you are building and where it has to run.
Talk to us