About the client
Client’s challenge
AI offered real practical value for the company. It could take over repetitive yet critical tasks, such as generating reports, reviewing data for inconsistencies, and assisting clinicians with quick, data-based decisions. The goal was to reduce manual effort and improve the system’s accuracy and usability in real clinical settings.
The company needed engineering support to introduce AI into its product. Building a reliable model and integrating it into a regulated EMR environment required specific skills in data science, backend architecture, and compliance. Recruiting such specialists locally proved difficult and time-consuming. To meet release deadlines and control costs, the company decided to bring in an external engineering partner.
- No in-house expertise in applied AI
- Patient data stored in both forms and free-text notes
- Heavy manual effort in preparing and reviewing reports
- New AI features needed to work inside existing clinician workflows
- Need to maintain HIPAA / HITECH compliance
Solution
Architecture and stack
We used PyTorch to design and train the AI models behind report automation and data analysis. After training, we deployed the models on Vertex AI within the client’s Google Cloud environment, where they automatically adjust the amount of processing power based on load.
Data flow and integration
Integrating AI into an existing EMR was not a plug-and-play task. Data formats differed across modules, and some were not suitable for direct model input. Our engineers tested several options for preparing and routing the data — from using intermediate buffers to direct API calls — before choosing a direct, API-based setup with in-memory preprocessing. This approach was fast enough for executing real-time reporting and fit into the EMR’s existing security and compliance environment. They also adjusted the preprocessing logic to handle uncommon or irregular data patterns in clinical notes, like typos in doctors’ notes or terminology misuse.
Compliance and security
We also made sure the new AI functions followed HIPAA and HITECH rules. The automation works entirely inside the EMR’s secure environment: the module processes de-identified data through encrypted APIs; each operation uses role-based access control linked to EMR user permissions; all AI requests are logged for audits, including user, time, and data type; no patient information is stored or cached outside the EMR.
User interface integration
On the front end, we added new functionality to the existing React interface, enabling clinicians to create AI-assisted reports and view alerts without needing to open another app, switch tabs, or change their workflow in any other way.
AI-assisted reporting data flow
within the EMR
clinical notes
findings identification
(Report builder)
visible in workflow
Process
1. Requirements alignment
2. Prototyping and validation
3. Project management routine
1. Requirements alignment
2. Prototyping and validation
3. Project management routine
“We could tell we worked with experts. NEKLO team spoke our language, gave helpful advice on tech architecture and Al-driven models, and seemed to genuinely care about the result. We felt like we were in very good hands.”
“The NEKLO team helped us meet our release deadlines and made sure the clinical transformation process was smooth, effective, and affordable.”
Technologies
Front-end
Back-end
AI/Data
DevOps / Infrastructure
.svg)
Deliverables
.png)
