EmbryoNet — Case Study

EmbryoNet AI Technologies

Science, translated into product

I helped turn a validated AI and microscopy technology into a laboratory-facing software product. My work ran from product vision and UX architecture to prototypes, delivery, and client-specific workflows.

CPO / Product Design Lead Scientific AI Automated microscopy Lab workflows

Hero visual: microscope or product shot. Screens can be blurred or reconstructed for confidentiality.

1 yractive product work
+1 yrsupport & customization
4major product versions
~50screens with states
3developers, front & back

The science was ready. The product was not.

I had been close to the project for years, helping with smaller visual and presentation tasks. In 2024 I joined as a core product contributor, when the research, publications, AI models, and computer-vision foundation were already strong.

What was missing was the product layer. There was no laboratory-facing software, no interface for interacting with the microscope, no website, and no coherent product narrative for demos, conferences, and potential clients.

The uncertainty was never whether the technology worked. It was which industry pain to lead with, who the primary users and buyers were, and how to turn scientific capability into a product people could understand, test, and trust.

Users & workflow

Designing around the experiment lifecycle

The users were scientists in small and mid-sized laboratories, plus research teams inside a large pharmaceutical client. Their core task was to test new drugs and compare which experimental conditions looked most promising.

Before touching the software, they prepared samples, chose experiment conditions, placed samples into plates, and defined observation parameters. The interface had to mirror this physical laboratory process instead of forcing it into generic dashboard logic.

Experiment userflow, sanitized reconstruction

lab process → software → decision points → output
01
Lab preparationOutside the software
02
Project setupCreate experiment project
03
Samples & conditionsPlates, conditions, parameters
Setup valid?Validation gate
No Fix configuration and revalidate ↺ back to 02
04
Send to microscopeJob dispatched to hardware
05
Microscope runAutomated observation
06
AI analysisSegmentation & phenotyping
Deviations detected?Model check
Yes Flag and annotate for scientist review
07
Compare conditionsRank the most promising results
08
Report / outputDecision-ready results

Not one screen. A working system.

The product spanned the full experiment: setup before the run, interpretation during and after, and preparation of results for real decisions.

Setup01

Prepare the experiment

Onboarding, project creation, sample setup, experiment conditions, microscope configuration, admin and settings.

Reconstructed setup / project-creation flow
Analysis02

Interpret time-series data

Dashboard, image and video analysis, time-series viewer, model segmentation, deviation detection, and comparison views.

Blurred analysis / segmentation screen
Output03

Move toward decisions

Reports, statuses, alerts, error states, client-specific workflows, and preparation of results for downstream use.

Report / results output
Product screen, blurred or cropped

Actual product screen. Blur or crop as needed for confidentiality.

Sanitized reconstruction

Reconstruction showing the logic without exposing sensitive detail.

Key decisions

Designing for trust and control

1

Make AI a working analysis tool

Scientists needed to trust AI-assisted analysis inside a serious research workflow. I structured the AI experience around visible analytical steps: time sequence, segmentation, detected deviations, comparison, and report preparation.

2

Separate setup from analysis

Before the experiment, users needed control and validation. After it, they needed comparison and interpretation. I split the product into distinct setup and analysis workspaces.

3

Show system states clearly

In a hardware-connected workflow, users need to know what is configured, what was sent to the microscope, what is running, what failed, and what is ready for analysis.

4

Support client-specific workflows

I designed a core workflow flexible enough to adapt selected processes for specific clients, without rebuilding the product from scratch each time.

DraftConfiguration still incomplete
Ready to runAll required setup is valid
Sent to microscopeExperiment data submitted
RunningObservation in progress
ProcessingAnalysis is being prepared
Analysis readyResults can be reviewed
Incomplete dataUser attention required
ErrorCorrection or retry needed

Leadership & delivery

A small team, fast rhythm, real clients

I led a product team of four: myself as CPO and Product Design Lead, plus three developers across frontend and backend. My job was to translate scientific and client needs into product decisions, prototypes, development tasks, and iterative releases.

ImmersionScientific context, lab process, and client workflow observation
RequirementsTechnical specs, product tasks, and feature logic
PrototypeInteractive prototypes for each major feature
DevelopmentFigma, Trello, daily calls, and informal developer communication
ReviewBiweekly feature reviews prepared for the wider team
IterationClient feedback, product revisions, and custom workflow adaptation
Scientists
Deep immersion at the start, then focused calls whenever a feature needed scientific clarification.
Clients
About a week observing the current process, a week designing, then roughly two iterations before development.
Developers
Prototypes as shared specifications, Trello tasks, daily calls, and AI-assisted tools for faster exploration.

Results

From validated method to working product

The project moved from a scientifically validated technology toward a working software product that could be demonstrated, tested, customized, and used with real partners and clients.

The outcome was confidence. Users could follow the process, read system states, and trust the interface while moving from manual lab work to automated, AI-assisted analysis.
91%model accuracy on validated benchmarks
10×faster than manual analysis
2processing modes: real-time and batch

Reflection

Product design here was translation: from science to software, from laboratory practice to interface logic, from AI capability to a workflow people can trust.

The most meaningful part was helping build a tool that lets researchers study complex biological processes faster, and with more confidence.