Amir Golparvar.

I joined NOMAD in 2020 as a backend developer and grew into full-stack ownership over four years. NOMAD is a federated platform for materials science research data: it ingests, validates, processes, and makes searchable over 19 million entries and 114 TB+ of scientific data across 10+ partner institutions, serving a community of 600+ researchers.

On the backend, I design and operate FastAPI microservices for ingestion and processing, orchestrate distributed task queues with Celery and Temporal, and manage the Elasticsearch indices that drive faceted search across heterogeneous metadata from 60+ scientific file formats. On the frontend, I lead the NOMAD GUI — React + Vite + Material UI + TanStack Query + Plotly — for uploads, search, interactive data viewers, and admin tooling.

The platform runs in two deployment modes: single-node Docker Compose for partner institutes who run their own instance, and full Kubernetes (Helm) on Max Planck's HPC infrastructure at MPCDF Garching for the central production deployment. I own the deployment story for both, including orchestration of Elasticsearch, MongoDB, RabbitMQ, Keycloak, and nginx behind a reverse proxy.

I also develop NOMAD plugins — parsers and schema packages that extend the platform to new scientific domains — and co-developed the Electronic Lab Notebook (ELN) module that lets researchers capture and share experimental workflows. The full codebase ships under TDD/BDD practice: end-to-end Playwright suites, integration tests with pytest, module docs in MkDocs and Storybook.

NOMAD's surface area is enormous — schemas, plugins, parsers, deployment patterns, the metainfo system, dozens of integration paths — and onboarding used to mean a week of reading scattered documentation and tracing example plugins by hand.

I architected NOMAD Compass: an internal RAG assistant that ingests the full NOMAD documentation tree, schema definitions, and selected example code, and answers natural-language questions with explicit citations back to source. The retrieval pipeline embeds every documentation chunk and code example into a vector store, and a Python service handles query, context selection, and response generation with source attribution. What used to take a week now takes a conversation.

FoamPrisma is a self-hosted AWS deployment that rebrands and extends NOMAD for OpenFOAM CFD simulation data — the same architectural ideas applied to a different scientific domain. It treats nomad-FAIR as a versioned dependency rather than a fork, following the nomad-distro-template pattern: a six-repository structure where upstream changes flow in via automated daily rebases, while CFD-specific extensions (an OpenFOAM v2206 parser, Temporal workflows for long-running simulation ingest, S3 storage adapters) live in their own packages.

The constraint that drove the architecture: the internal Python package retains the name nomad to facilitate clean upstream merges, while every external surface — GUI, API, branding — adopts the FoamPrisma identity. It runs in two GUI modes (/gui-v1, /gui-v2) to support both legacy and new workflows during transition.

P3D-BRNS is a three-dimensional, multiphase, multicomponent, pore-scale reactive transport modelling package for simulating biogeochemical processes in subsurface environments. It takes micro-CT images of porous structures — soil, rock, biofilm — discretizes the pore space, and solves the coupled flow, transport, and reaction equations directly on the geometry.

I built it during my PhD work as a research toolbox; the v1.0.0 release in 2024 was the formalization, published as a software paper. The codebase is C++ on top of OpenFOAM, with the reactive transport solver coupled to OpenFOAM's finite-volume CFD machinery. Open source and used by other research groups working on subsurface biogeochemistry.