Self-taught software engineer. I build LLM-powered products and agent systems end to end, with AI as the method — orchestrating several agents as a one-person team. Current interest: medical software and fintech / payments. TypeScript + Python. Open to remote roles and relocation. In Deutschland ansässig; Deutsch und Englisch, remote wie vor Ort.

📫 cedric@condaydigital.com · LinkedIn

xe-mcp — MCP server for the Xe Currency Data API (TypeScript · 49 tests · CI · AWS Lambda) FX tools — rate, conversion, volatility, optimal send-window, correlation, charts — with a free ECB fallback so it runs with zero credentials.

centrapay-mcp — MCP server for Centrapay, NZ payments (TypeScript · npm · contract tests) Nine tools modelled on the real Centrapay API — payment requests, settlement, idempotent refunds, merchants — with fetch-mock contract tests over every request.

Protocol Tracker — clinical-grade health app for a real treatment protocol, shipped in two builds:

• v1 — cross-platform native (React Native · Expo · TypeScript · SQLite): multi-provider LLM vision-OCR (Anthropic / OpenAI / Groq) reads a photographed medical report and auto-fills structured data — BYO-key, with token & cost guardrails; offline-first SQLite (versioned migrations), biometric auth, PDF reports, EN/DE, accessibility-oriented.
• v2 — re-architected as a zero-knowledge PWA (vanilla JS · Capacitor iOS/Android · IndexedDB · PBKDF2→AES-GCM-256): all patient data encrypted client-side, no server, encrypted backup/restore — privacy by design shrinks the compliance surface to zero.

nifti-qc — quality-control for medical-imaging geometry (Python) Catches silently-broken NIfTI qform/sform geometry before it corrupts a neuroimaging pipeline — the kind of quiet data bug that invalidates a lesion analysis without ever raising an error.

Bugs found, reproduced, fixed with a regression test, and merged upstream — 39 merged across 16 established projects:

Medical & neuro-imaging

• mne-python — eyetracking calibration files read as UTF-8 (neuroimaging)
• nibabel — set patient_birth_date dtype in the ECAT header to signed int32 (neuroimaging — also earned a CITATION.cff credit)
• In review: medical-imaging campaign — dicom2nifti, LST-AI (DICOM / NIfTI · MS lesion tooling)

Fintech & data

• schwifty — correct ISO 7064 mod-97-10 checksum for BA (IBAN)
• faker — compute a valid ISO 7064 Mod 11,10 check digit for the de_DE vat_id (test-data)
• aeon — Padder validates fill_value, rejects array-likes (time-series ML)
• dpdata — build pymatgen Molecule species from atom_types order (computational materials · DeepModeling)

Developer tooling & infra

• jaeger — MCP tooling reports total_count + truncation (observability · MCP · CNCF)
• octokit/request.js — handle primitive JSON error bodies without throwing (GitHub SDK)
• js-joda — Instant.parse accepts and resolves an offset (date-time)
• krkn-ai — include dependency in CompositeScenario identity (chaos engineering)

More merged across DuckDB, growthbook, django-scim2, and Adyen; further fixes in review across OpenTelemetry and cargo-deny.