Case Studies

Context

A clinical voice analysis platform serving speech-language pathologists. The platform needed to replace manual Praat/Excel workflows with real-time automated analysis, covering 170+ validated acoustic measures across multiple clinical protocols.

Challenge

• No existing open-source platform covered the AVQI, DSI, CPP, jitter/shimmer, formant, and prosody measures needed
• Clinical data sensitivity — GDPR compliance and Swiss data residency required
• Multiple stakeholder types: clinicians, patients, researchers, administrators
• Real-time waveform and spectrogram visualization at clinical quality
• Integration with clinical databases and export to standard formats

Approach

Built a full-stack SaaS platform on Azure with OVHcloud-Switzerland storage for data residency. The audio processing pipeline was built on Python/FastAPI, with a React frontend. The DSP layer implemented 170+ acoustic measures using librosa, Praat via Parselmouth, and custom C/WASM modules for real-time performance.

ML components included: Wav2Vec2-based pronunciation assessment, eGeMAPS feature extraction via openSMILE, speaker verification, and pathology classification models trained on clinical datasets. A multi-agent AI system generated personalized therapy plan suggestions.

Outcomes

Tech stack

Context

An ERP application suite for a railway alliance bringing together several co-contracting companies. The foundation needed to replace scattered document exchanges with a single flow — contracts, purchase orders, receptions, and invoices — shared across teams from different companies.

Challenge

• Several co-contracting companies needed to share the same contract, purchase order, and invoice repository without duplicating data or access
• Purchase → reception → invoice chain previously handled manually, a source of errors and delays
• Contractual document signing needed to go electronic, with legal evidentiary value, without falling back to paper
• Field teams' hours and travel allowance tracking lacked a centralized tool
• User accounts and access rights scattered across applications, with no single source of truth

Approach

Architecture built as dedicated applications, each responsible for one business domain (contracts and purchasing, electronic signature, timesheets), connected by a shared identity service (single sign-on, JWT tokens) and a centralized user management portal. A shared UI component library ensures a consistent experience and avoids rebuilding the same screens in every application.

Outcomes

Tech stack