WalidBoudabbous

Chief Technology Officer AI & Healthcare Platforms

Twenty-eight years of delivery accountability across healthcare, neobanking and large-scale e-commerce. Today: a 22-person engineering organisation and a clinical platform that runs entirely on its owner's hardware.

Hamburg, Germany w.boudabbous@dudoxx.com LinkedIn GitHub

Chapter II What I run

One clinical core where a clinic used to run five to eight disconnected tools.

As Chief Technology Officer of Dudoxx UG I own technology strategy, architecture and delivery for the Dudoxx FHIR Clinic Platform, leading a 22-person cross-functional organisation across frontend, backend, AI/LLM, design, product and QA.

The platform replaces a stack of EHR, scheduling, billing, dictation, document AI and patient portal with one FHIR R4 clinical core, one identity model and one audit trail. Deployment is a configuration choice, not a fork: a managed EU cloud service and a fully customer-operated on-premises install share one application core, so data residency and self-hosted inference are selected rather than rebuilt.

22 engineers across six disciplines, one organisation
11 roles, surfaced through 14 role-scoped portals, specialty-configurable
~1,470 REST endpoints across ten domains
184 domain data models behind seven deployable capability areas
380 spec and test files gating continuous integration
0 external LLM calls, by architecture

Chapter III The decision

Every model runs on our own hardware, so no patient record ever leaves the building. It came out faster and cheaper anyway.

Choosing a fully on-premise architecture with zero external LLM calls was a data-sovereignty decision, and it was the expensive-looking one at the time. It holds AI response under 800 ms and cuts inference cost five to eight times against cloud list price.

That posture is enforced in the request path, not in a policy document. Three tenant-isolation layers run from request interceptor through database row filtering to a per-clinic FHIR partition, extended into object and vector storage. RBAC is a server-side boundary the interface cannot weaken. An append-only audit log records every call. PII is redacted at the LLM egress gateway, which makes the AI boundary itself a compliance control.

The AI estate is governed the same way: an orchestrator spanning 233 tools across 32 domain categories calls the same RBAC-enforced endpoints the interface calls. No agent holds privilege the user does not, and there is no privileged backdoor for the model.

Chapter IV The ledger

What shipped, and what is still in beta.

Engineering pages usually list only the wins. This one carries the status of each capability as the platform's own architecture record states it. Anything still planned is absent entirely.

FHIR R4 clinical core. Twelve resource types behind a resource facade over HAPI FHIR, with a per-clinic partition per tenant.
Shipped
Encounter lifecycle, end to end. Check-in, call-next, enter-room, complete, wired through the clinical record.
Shipped
Scheduling. Templates, recurrence, availability, slots, waitlist and public online booking across 31 controllers.
Shipped
Revenue cycle. Claim through ClaimResponse, adjudication and payment reconciliation, with GKV, PKV and self-pay modelled separately.
Shipped
Double-entry accounting. VAT, fiscal-period close, profit and loss, balance sheet, general ledger, trial balance, dunning and AR escalation.
Shipped
Telemedicine. In-platform video on infrastructure and keys the clinic owns, with no per-visit fee.
Shipped
Immunisation and prescription flows. Completeness, recommendations, overdue tracking, compliance reporting and repeat-count renewal chains.
Shipped
Trusted gateway. The browser never reaches the API directly; a backend-for-frontend re-signs every request, so API keys never leave the server.
Shipped
Append-only audit trail. Every call recorded against the calling identity, behind a fixed guard chain.
Shipped
Agentic orchestration. A successor engine lowers one structured plan per turn into a bounded execution graph; mutating actions are never retried and never taken without a typed human-approval gate.
Shipped
Real-time voice plane. Ambient consultation capture to structured transcript to a clinical note written back into the record.
Beta
Document intelligence. Optical character recognition into structured clinical observations, and retrieval-augmented answering over the clinical knowledge base.
Beta
Billing automation. Automated coding assistance on top of the shipped revenue-cycle core.
Beta

Two figures I hold as design targets rather than results: the voice intake flow is designed to take patient intake from fifteen minutes to under a minute, and to reduce provider documentation time materially. Both are engineering intent. Neither is measured in production yet, so neither is claimed as achieved.

Chapter V Twenty-eight years

Healthcare, neobanking, and e-commerce at global scale.

2023 – PresentHamburg, Germany

Chief Technology Officer, Dudoxx UG

22-person cross-functional organisation.

Own technology strategy, architecture and delivery for the Dudoxx FHIR Clinic Platform, an AI-first healthcare management system with telemedicine.
Made new clinics and indications onboard without engineering: administrators author versioned intake and assessment forms from a portal builder, and clinical workflows publish as data through a transactional draft, publish and archive lifecycle.
Consolidated four products onto one shared AI substrate, cutting duplicated build and run cost. Clinical SOAP notes draft in under eight seconds.
Architected DDX Omni-Intelligence, the self-hosted CUDA inference platform the products run on: own speech-to-text, text-to-speech and LLM serving behind one OpenAI-compatible gateway, with CUDA and Apple-Silicon backends sharing a single vendor-neutral wire envelope.
Set the delivery quality bar: a single-command environment bootstrap that puts a new engineer under ten minutes from clone to running stack.

2024 – PresentHamburg, Germany

Chief Technology Officer, Derm4hair

Own the technology agenda for the clinic, setting the AI-assisted videotrichoscopy and hybrid telemedicine architecture behind its hair-health and regenerative-dermatology service lines.

2016 – PresentTallinn / Hamburg

Principal Technology Consultant, Acceleate Consulting OÜ

Independent practice. CTO-level accountability for digital-health and fintech platforms, from board-facing strategy through delivery.

Shae Group / Anthrophi Technologies (2026) – gave the board a decision-ready verdict on a large production platform: a code-grounded architecture assessment framed as 19 architecture decision records and a phased rewrite programme.
Weeca Neobank (2021–2023) – Project Director and CTO. Owned the ground-up build and market launch of a neobanking platform for Azur Services GmbH over two years, from licence-facing architecture to live operations.
Clinaddict, Cannes (2025–2026) – accountable for the patient platform and clinic operations system of an addiction-treatment centre.
Direct Medical Access, Dubai (2023–2024) – directed an AI-driven healthcare portal and a medical-evacuation patient-record platform.
Opened the Tomedo EMR integration route for German medical practices and stood up their self-hosted open-source LLM deployments (2023–2024).

2006 – 2015Tunis, Tunisia

Founder & Managing Director, Acceleate

Grew an offshore delivery organisation to 48 full-time engineers serving European enterprise clients, owning profit and loss, hiring and client relationships.
Built the operating model that made offshore delivery sellable to European buyers: a repeatable method for standing up Agile teams and skill centres at European quality levels.
Held delivery accountability for www.louisvuitton.com on behalf of IBM over three years.

2013 – 2015Paris, France

Senior Delivery Manager, Wipro France

Accountable for Digital Service Line delivery across Continental Europe. Reset delivery process and team performance across the portfolio.

1996 – 2013Europe

Earlier roles

Senior Oracle Architect, Wipro NL and Philips, Amsterdam (2012–2013) – owned long-term e-commerce strategy for the Philips group and won the case for consolidating every ATG implementation across the group onto one platform.
Technical Manager & Solution Architect, Acceleate, Tunis (2010–2012) – delivered Zitouna Bank's internet and mobile banking on TEMENOS core banking, secure, real-time and fully bespoke.
Business Unit Manager, ATG (Oracle), Paris (2001–2006) – ran the France business unit with revenue accountability; owned the Orange Global Web Self-Service Platform architecture and the Louis Vuitton e-commerce platform across Europe, Japan and the USA.
CRM Project Manager, Syntegra / British Telecom, Paris (1998–2001) – created Syntegra's Integration Department through a management buyout programme, building it to 36 engineers.
Research Engineer, École Centrale de Lyon (1996–1998) – numerical models for mechanical friction and metal wear; taught at ENISE.

Healthcare & AI platforms

HAPI FHIR 8.4, FHIR R4/R5, telemedicine, HIPAA-aligned controls, LLM gateway and routing, retrieval-augmented generation, on-premise inference architecture, self-built CUDA speech-to-text and text-to-speech, vector search, prompt engineering.

Security & platform

Zero-trust architecture, RBAC, JWT, PII redaction, audit trail, AWS, Docker, Kubernetes, Jenkins, CI/CD, microservices, MinIO, LiveKit, PostgreSQL 16, Redis 7.

Engineering & delivery

Java/J2EE, Python, Node.js, NestJS, FastAPI, Prisma, React 19, Next.js 16, TypeScript, Agile, Scrum, Kanban, offshore delivery management, Jira and Confluence.

Education

École Centrale de Lyon: Engineering degree, Mechanics & Computer Science, 1996
DEA, Geomaterials & Solid Mechanics: École Centrale de Lyon & INSA Lyon, 1996
Languages: French native, English full professional, German limited working, Italian beginner