Arperon is a small software engineering studio from Slovakia. We ship products, run an applied AI research track, and take on selective custom engineering — for domains where the software is not allowed to be wrong.
Arperon is a parent studio. Everything we do falls into one of three tracks, and every track feeds the others. ThrustCore is our most visible output, but it isn't the whole picture.
We ship our own software. ThrustCore is the flagship — a FHIR R4-native clinical platform designed around Indonesia's SATUSEHAT framework, and usable as a general clinical information system anywhere else. More products are in early stages.
We maintain an active applied research track on agentic AI systems — tool-calling loops, model-context protocol servers, multi-provider parity, and safety for clinical use. Research outputs feed our products and client work.
We take a small number of selective engagements each year, in domains where we already go deep — domain-heavy web apps, AI integrations, healthcare interoperability, and performance-critical PostgreSQL work.
Our flagship. A modern clinical information system in which every clinical fact — a diagnosis, a prescription, an encounter — is stored as a standard FHIR R4 resource from the moment it is recorded.
ThrustCore was designed specifically to help hospitals, clinics, and care networks in Indonesia meet the requirements of the Ministry of Health's SATUSEHAT interoperability framework — and remain an excellent clinical information system anywhere else.
Because FHIR is the architectural substrate and not a post-hoc export format, standards compliance is a property of the system, not an expensive retrofit at the end.
The platform ships with a patient mobile app, a staff web GUI, a Supabase-backed API, and an MCP server for AI integrations — all wired together through a single FHIR R4 API, SMART-on-FHIR authentication, and row-level security at the database layer.
Request the platform brief →Dozens of resource types — Patient, Encounter, Observation, Condition, Procedure, MedicationRequest, CareTeam, Consent, and more. Dual-coded conditions (ICD-10 + Kemenkes clinical-term) for SATUSEHAT alignment, with country-aware terminology resolution.
An agentic loop with real FHIR tools: search conditions, observations, medications, encounters. Provider-portable across Anthropic, OpenAI, OpenRouter, and Gemini tool-calling formats.
Real-time messaging between patients and whole care teams. Per-user read status, templates, bulk messaging, medication requests.
Every stock movement becomes a standards-compliant EPCIS event — traceable, auditable, and exportable end-to-end.
Labor rules are JSON-configured per country and organisation — never hardcoded. Rosters, holidays, time-clock, absences.
A real-time queue and scheduling module designed around actual physiotherapy department workflows. Deliberately small — the right entry point for a single-department deployment, and a proving ground before the rest of the platform rolls out.
Row-level security at the database layer. Organization switching at the token layer. National organizations own shared code systems (ICD-10, ICD-9-CM, SNOMED CT specialties) — every tenant sees the same source of truth.
We treat AI as engineering, not magic. Our research track focuses on agentic systems and tool-calling — the parts of AI that have to actually be right when deployed next to a clinician.
We don't build toy demos. Every AI system we ship has to make real API calls against real clinical data — and stop when it cannot do so safely.
Our research focuses on problems that are easy to hand-wave and hard to ship: how does an agentic system know when to stop? How do you bridge tool-calling formats across providers without rewriting your prompt every six months? How do you audit what an AI assistant actually did during a consultation?
The output of this track is part-product, part-paper. Results feed directly into ThrustCore and into the custom engagements we take on.
Replacing pre-fetched context with agentic FHIR tool calls inside a production clinical copilot. Bounded iterations, soft timeouts, and honest failure modes.
A single tool surface compiled into four providers' tool-call formats — so a clinical copilot stays portable as the provider landscape shifts.
Using the Model Context Protocol to expose narrow, auditable slices of a FHIR system to AI assistants — without leaking more than the task requires.
Every tool call, every retrieval, every response — recorded as structured, signed events. How you make an AI copilot legible to the people who have to sign for it.
We take standards seriously because they are how clinical systems become honest. Everything below is implemented in ThrustCore today.
Architecturally ready for Indonesia's national health data exchange. Dual-coded diagnoses, Kemenkes clinical-term catalog, Bahasa Indonesia display text, country-aware terminology, and encounter diagnosis coding.
ES256-signed JWT tokens, organisation-scoped claims, role and privilege arrays resolved at login, a unified role system across patients and staff, and an immutable audit log underneath.
Built with Europe's GDPR and Indonesia's Personal Data Protection Act (UU PDP, 2022) in mind. Fine-grained consent tracking as FHIR Consent resources. Deployable to Indonesian regional data centres for residency compliance.
National organisations own the code systems. All tenants in the same country read the same source of truth. ICD-10 diagnoses for ID and SK, ICD-9-CM procedures, SNOMED CT specialties — each translated to EN, SK, ID where applicable.
A deliberately narrow stack we can carry across a ten-year product lifecycle without drama. Depth over breadth, every time.
Healthcare software is stubborn work. The domain is messy, the standards are heavy, the consequences of bugs are real. We pick a stack we can still love in five years.
Our engineering values are simple: FHIR from day one, types everywhere, minimal abstractions, root-causes not bypasses, and a refusal to paper over bad architecture with feature flags. We read the migration before we write it.
Outside ThrustCore we take on a few custom engagements each year — always in domains where we already go deep. Healthcare, AI integration, domain-heavy web apps, PostgreSQL performance work.
| Frontend | Next.js 15 · React 19 |
| Mobile | React Native · Expo SDK 53 |
| Backend | Supabase · PostgreSQL · Edge Functions |
| Language | TypeScript · SQL · PL/pgSQL |
| Standards | HL7 FHIR R4 · SMART · GS1 EPCIS 2.0 |
| AI | Anthropic · OpenAI · Gemini · tool-calling loops |
| Auth | ES256 JWT · RLS · unified role system |
| Integrations | Model Context Protocol · custom MCP servers |
Whether you have a clinical system to modernise, an AI integration to get right, or a hard software problem in a serious domain — we read every message.
