I'm Elnaz Alikarami — I build systems that make scientific and clinical data trustworthy.
I came to software from neuroscience. Human brain imaging taught me how research data actually goes wrong: fragmented across incompatible systems, cleaned by hand, corrected in ways nobody can reconstruct six months later. I now build the infrastructure that stops that happening — with the domain knowledge to know which errors matter and which are noise.
Pronouns: she/her
Clinical Data Fabric System — a clinical data platform: multi-source ingestion, CDISC-aligned standardisation, validation, correction with full downstream propagation, and submission-ready SDTM/ADaM export where every row traces back to what it was built from.
Built solo in Python. Zero third-party runtime dependencies, 677 tests at 92% branch coverage, CI on 3.11–3.13. That repository is a public summary; the source is available on request.
N-DOS — a system for recovering, describing and querying animal neuroscience data that nobody organised. Point it at a drive whose collector has graduated: it reads what is there without changing it, reads inside the archives without extracting them, and rebuilds a usable layout — subject, session, data type — from whatever structure already exists, labelling every inference as an inference so you can correct it rather than trust it.
Then the parts that make it worth keeping. Full-text search across filenames, lab notes, protocols and spreadsheets — including Word and Excel — where a hit in a surgery log names the animals it mentions and points at their recording sessions. Cohort queries that return three answers rather than two: matched, excluded, and cannot be ruled out, because a session whose species nobody recorded is not a session known not to be a mouse, and collapsing those two cases biases a cohort quietly. Provenance from a figure back to the raw files behind it. Handoff to BIDS and NWB. An optional local interface, for the people in a lab who do not work at a command line.
Built solo in Python and released: pip install ndos. Zero third-party runtime
dependencies — it has to run on locked-down acquisition machines, and CI proves it by
running every module against a bare interpreter — with 444 tests on 3.9 and 3.13 across
Linux, macOS and Windows. Designed against real lab drives rather than synthetic
examples, which is where most of its design came from: the discovery that most of the
data on an inherited drive is sitting inside archives nothing had ever opened changed
the whole approach.
Looking for labs to pilot it. If you have a directory nobody fully understands any more, that is exactly what it needs to meet — it takes about fifteen minutes and will not move or change your data. Start here.
Alongside both: data harmonisation and reproducibility across multi-lab settings, and research data management in neuroscience labs.
Engineering — Python, SQL, REST API design, schema design, pytest, CI/CD, Git, Docker
Data — pandas, NumPy, data modelling, lineage and provenance, pipeline design
Clinical & research standards — CDISC (SDTM, ADaM), Define-XML, UCUM, 21 CFR Part 11, HIPAA/GDPR/PIPEDA handling of identifiers
Research — fMRI and neuroimaging analysis, experimental design, R, MATLAB
- Designing data systems that survive an audit — lineage, provenance, and why "we fixed it in the spreadsheet" is a problem you find out about much later
- Running a research lab like a startup 🚀
- Brain imaging, cognitive neuroscience, and behavioural data
- Bridging the gap between the people who collect scientific data and the people who build systems for it
Still mapping systems, just more than only neural ones. Same goal: make sense of the noise.