AI and integration engineer in Bergen, Norway.
I build backend integrations and AI application-layer systems that run in production. Three years at ID WRKS, a Bergen IT company, went into the automation layer on top of Halo PSA: a retrieval-augmented support workflow that pulls similar past tickets from a vector store, rebuilds them into conversation form, and sends that context to Azure OpenAI for a first-line answer, with runbook filtering before anything reaches the model. On the same platform I automated CSP license invoicing end to end. A recurring run that took over three hours now takes under ten minutes, and the manual reconciliation path went with it. I repeated the pattern for the N-able customer base. Also from those years: an HPE Aruba Central integration that syncs switches and access points into Halo PSA as managed assets with unattended OAuth token refresh, an e-signing integration built on per-document webhook callbacks instead of polling, Microsoft Entra ID administration, and dynamic T-SQL across a 14,000-column schema.
The stack is Python-led: FastAPI, REST and webhook integrations, SQL and T-SQL, Docker and Docker Compose behind Traefik, GitHub Actions, pytest, Linux. On the cloud and identity side, Azure (VMs, Functions, Azure OpenAI) and Microsoft Entra ID, including RBAC, app registrations and OAuth 2.0 client-credentials flows. On the AI side, RAG in production, MCP servers, the Anthropic API, LangGraph and LangChain, n8n, and prompt engineering as daily work. Earlier, during PhD research at NTNU from 2020 to 2023, I worked in TensorFlow, Keras and PyTorch, fused acoustic and radio-frequency signals for in-home activity recognition and indoor localisation, and deployed CNN models down to a Raspberry Pi. I left NTNU before defending. The signal-processing foundation is still the part I use most.
Most of that work lives in employer production systems or private repositories, which is why this profile currently shows research-era forks and little else. I am changing that. The first repository I am preparing to publish is JobPipeline, an agent system that runs my own job search end to end: 17 scheduled Claude Code routines covering discovery, scoring, CV and cover-letter drafting, a Telegram approval flow, and compliance reporting. Each routine is pinned to the cheapest model that can do its job (Haiku 4.5, Sonnet 5, Opus 5) and runs under its own token budget, so the cost of the automation stays visible. My homelab at masoud-mh.com runs on a headless Ubuntu server I set up myself, with Docker Compose services behind Traefik and Let's Encrypt certificates.
- Homelab: https://masoud-mh.com
- LinkedIn: https://linkedin.com/in/masoud-mohtadifar-5b6b758a
- Email: masoud.mohtadifar@gmail.com
- Based in Bergen. Open to roles anywhere in Norway and to fully remote work.