A precision-nutrition decision platform for one of medicine's most unforgiving environments.
We built an AI-driven feeding-recommendation system for the Medical Client — turning a proprietary neonatal dataset into protocol-backed guidance that nurses can act on, in real time, at the bedside.
Client's data was scientific. Their software was nonexistent.
The Client is a US-based precision-nutrition company focused on outcomes for preterm infants. Their clinical work had produced a rich, proprietary dataset — feeding protocols, microbiome profiles and longitudinal growth metrics — one of the most comprehensive neonatal nutrition datasets in existence.
The problem: that dataset lived in disconnected systems. Clinical teams couldn't act on it in real time. Feeding decisions were still made manually — from memory and judgement — without systematic access to what the data suggested. Astarte needed a software partner that could turn the science into a clinical tool.
Preterm care is unforgiving. The cognitive load is enormous.
Feeding decisions affect gut development, growth velocity and long-term neurological outcomes — and must be made multiple times daily by nurses managing several critically ill patients at once. The constraints shaped every part of the build.
Zero — no prototype, no legacy system to extend. The platform had to be designed and built end-to-end, from data model to clinical UI.
Recommendations had to integrate historical intake, current growth curves, microbiome indicators and protocol parameters — all varying per patient.
The system had to augment clinical judgement, not replace it. Every AI suggestion needed to be traceable, with an override path and a logged reason.
Bidirectional connection with Epic — patient records flowing in, compliance data flowing out — without disrupting hospital workflows.
A four-layer build, scoped to the clinical workflow.
We designed a PostgreSQL schema that brings feeding history, growth-velocity measurements, microbiome profiles and protocol parameters into a single queryable structure. Infant parameters are logged on admission and continuously updated through the care episode.
An AI-based decision tree processes current clinical indicators against Astarte's protocol library. It considers previous outcomes, growth speed, tolerance signals and contraindications to surface a prioritised recommendation — with the rationale attached so nurses can review and override if clinically indicated.
An Angular front-end presents recommendations in a low-friction interface. A dedicated rounding-report tab consolidates everything for ward rounds. Growth charts let leads review trends at individual and cohort level. Protocol modifications are logged and visible.
Bidirectional Epic integration keeps records in sync without manual re-entry. An analytics layer tracks compliance metrics across patients and time periods, giving clinical and research teams visibility into adherence and outcome correlations. All data is exportable for research and reporting.
Decision tree suggests timing, volume and constituents from clinical indicators and historical outcomes.
Visual tracking of growth velocity against expected curves, with deviation alerts.
Consolidated patient summary for ward rounds — all the relevant data, one screen.
Every modification logged with a reason code, enabling retrospective compliance analysis and audit.
Bidirectional sync — records flow in, compliance data flows out, no double-entry.
Role-based authentication and access control with audit trails for every data access.
Performance dashboard for clinical leads — adherence rates, feeding outcomes, protocol effectiveness.
Patient data and reports are exportable for external analysis and regulatory submission.
Astarte's proprietary microbiome profiles are first-class inputs into the recommendation engine.
Compliance went from untracked to measured. Decisions got faster. Nurses got their cognitive capacity back.
Compliance tracking surfaced consistent adherence across participating NICUs after platform adoption — a clear improvement on the pre-software baseline where deviations were untracked and common.
AI-generated recommendations reduced the time staff spent on feeding calculations, freeing capacity for direct patient care. Nurses report confidence in acting on system suggestions.
Bidirectional Epic integration eliminated manual data entry between systems, reducing transcription errors and keeping patient records current across the platform.
Data model, AI engine, clinical interface, EHR integration and analytics layer — all designed and delivered by our team, starting from no codebase.
“Client's proprietary dataset integrates feeding protocols, microbiome profiles and clinical information. We synchronised it with the application so it generates actionable suggestions in the form of a decision tree — giving nurses protocol-backed recommendations at the exact moment they need them.”
Built for reliability and clinical performance. Every piece of the stack was chosen for its track record in regulated healthcare deployments, not for novelty.
Building something for clinical teams?
Clinical decision support, neonatal care software, regulated AI systems — write to a principal engineer. We respond within one business day. No discovery call.