pimento / case studies

Case studies.
Real systems, real data.

Systems at our clients (medicine, research, compute infrastructure) and our own products. Each case study is a story page built on the real UI or on real, anonymised data: the scenes play by themselves, and in a few places you can take the controls.

Filter
R-01Fundacja MT5preview
01Signal overview · the whole nightOpen the story →
01Signal overview02The pipeline describes the night on its own03Close-up04Across many nights
Fundacja MT5deployed

A GPU cluster that learns to read sleep

The foundation's sleep labs record around 100 nights of studies a month. We supplied the hardware and a GPU cluster that sits at the client's site, built a pipeline that breaks a night down into events, stages and indices, and we train models on that data.

  • The story runs on a real, anonymised night (a diagnostic study)
  • Automated analysis: illustrative, not a physician's report
  • The model predicts a masked channel from the others
#medicine#AI#infrastructureNVIDIA BlackwellEDF+pipelinemodel trainingon-prem
View case study →
R-02Civilmedpreview
01Test queueOpen the story →
01Test queue02The slide under the magnifier03Semi-quantitative assessment and recommendations04Signing off the test
Civilmeddeployed

Cytology testing system

The journey of a single test: the lab technician scans the tube and counts cells on the slide, the physician assesses and signs off the result, and the system assembles a PDF. It runs on the client's server; the animation uses fictional data only.

#medicinemedical dataclient serverPDFaudit
View case study →
R-03Medical imagingpreview
01MPR 2×2: axial, coronal, sagittal, MIPOpen the story →
01MPR 2×20218.6 mm measurement and ROI033D VR04Report with measurements
Medical imagingdeployed

PACS with an in-browser DICOM Viewer

An imaging archive with a DICOM Viewer that runs in the browser: import into PACS, MPR 2×2, lesion measurement, an HU window computed live, 3D rendering of the lungs and a report that the measurements flow into on their own.

#medicine#infrastructureDICOMblob storageMPR3D VR

Imaging data: The Cancer Imaging Archive, LIDC-IDRI (Armato et al., Med Phys 2011), CC BY 3.0.

View case study →
P-01Dineezypreview
01The floor, liveOpen the story →
01The floor, live02Drag & drop floor layout03Timeline of the evening04Booking widget
Dineezyproduct

Reservations and a live floor plan

A reservation system for restaurants: a floor plan that updates itself, drag & drop table layout, a timeline of the evening, a new booking in a few clicks and a widget for guests on the venue's website.

  • Table statuses: free, booking coming up, guests seated
  • Guides and walkway checks for waiters
  • Guest widget on the phone
#productNext.jsPayloadreal timewidget
View product →
P-02sheptpreview
01Hold the key and speakOpen the story →
01Hold the key and speak02The message in the order you spoke03“shept” → agent → terminal04Insights computed locally
sheptproduct

Polish dictation and a voice agent

Hold a key, speak, let go: the text lands wherever your cursor is. Say “shept” and the command goes to the agent. Audio only ever goes to our own server.

#product#AImacOSspeech → textagentprivacy
View product →
P-03inferctlpreview
01Server fleet, liveOpen the story →
01Server fleet, live02Deploy with a command preview03Ansible in action04Chat with the model
inferctldeployed · in-house

Models on many GPUs from one terminal

The tool we run our own infrastructure with: a live GPU server fleet, model deploys with a command preview, Ansible playbooks and Terraform in one repo, and chat with a model straight from the terminal.

#infrastructure#AIPythonTextualAnsibleTerraformvLLM
View product →
coming soon
Pulse Vetcoming soon

Pulse for veterinary clinics

Our product for veterinary practices. We'll show it once it's ready to be shown.

#medicine#product#AIcoming soon
coming soon
Real-estate splatscoming soon

Interiors in 3D to explore in the browser

3D scans of flats and commercial spaces (Gaussian splatting) from a 360° camera, with a slow fly-through and rotation by finger. In progress.

#AI#productGaussian splattingcoming soon
Statuses: deployed = running (at a client or in-house), product = we sell it, coming soon = in progress. Previews are frames from the case-study animations; patient data is fictional or anonymised.See the live demo

Your project could be next.

We start with a conversation and a prototype on your examples. Your data stays with you from day one.