How to create a profile

A profile is a directory of static files conforming to the RP Format. The spec defines the format, not how to produce it. Any process that emits a conforming directory works, whether you write the files by hand, generate them with a script, or bootstrap them with an LLM.

Prerequisites

• Familiarity with JSON
• See the tutorial for a full walkthrough

What a profile directory looks like

At minimum, a profile is a directory containing two files:

jane-doe/
├── profile.jsonld
└── sources/papers.jsonld

profile.jsonld is the record and the manifest; sources/papers.jsonld is the works list, and it may be an empty Collection for a researcher with no publications. profile.jsonld lists it in hasPart with role works. A directory with those two files, where profile.jsonld carries conformsTo, is a conforming lite profile. Everything else is optional and additive:

jane-doe/
├── profile.jsonld
├── personality/
│   ├── expertise.md
│   └── SOUL.md
└── sources/
    ├── papers.jsonld
    └── summaries/
        └── <paper_id>.summary.md

personality/ and sources/ enable deeper features (persona methods, search, citation-backed answers) but are not required to have a valid profile. See the tutorial for a full walkthrough of building one of these directories by hand, file by file.

The required fields

profile.jsonld has three top-level keys with no default. Omit any one and the document fails to load (see the profile document spec):

Key	What it is
name	The researcher’s display name
rid	A canonical ORCID, or a local: id if they have none
provenance	Who asserted this profile and on what basis (no default)

Five more keys are required on the wire. Every conforming published document carries them, but the reference implementation fills them in when absent, so you can omit them while authoring by hand:

Key	What it is	Default when absent
@context	Exact string: https://profiles.databio.org/context/v1.jsonld	filled in with this value
@id	The subject IRI: https://orcid.org/<rid> for an ORCID rid, else the profile’s url, else #me	filled in from url or #me
@type	"Person"	filled in
conformsTo	Equals the @context IRI and acts as the format gate. A present but wrong value fails to load.	filled in with the current format IRI
level	"lite", "full", or "deep"	filled in as "full"

A minimal valid profile.jsonld:

{
  "@context": "https://profiles.databio.org/context/v1.jsonld",
  "@id": "https://orcid.org/0000-0002-1825-0097",
  "@type": "Person",
  "conformsTo": "https://profiles.databio.org/context/v1.jsonld",
  "name": "Jane Doe",
  "rid": "0000-0002-1825-0097",
  "provenance": "third_party",
  "level": "lite",
  "hasPart": [
    {
      "@type": "Collection",
      "name": "Works",
      "role": "works",
      "contentUrl": "sources/papers.jsonld",
      "encodingFormat": "application/ld+json"
    }
  ]
}

Everything else (personality documents, summaries, career history, collaborators) is optional. A lite profile with this document and the works list it names is conforming; it cannot run persona methods or feed a full-corpus search until it has more. rp manifest <dir> --write fills in hasPart from the files on disk, so you do not have to type it.

Ways to create a profile

• By hand: follow the tutorial step by step. It builds a small profile from scratch and explains each file.
• By script: any code that writes valid JSON (and, optionally, Markdown) into the directory works. There is no required tool; validate the output against the schema (see below).
• By LLM: paste the prompt below into ChatGPT, Claude, or any other model to generate a starter profile.jsonld.
• By pipeline: a profile directory can also be produced by an automated build tool. The format does not care which tool produced it; only the resulting directory is specified. The SDK defines and validates the format; it does not ship a build pipeline. See Creation is out of scope.

Generate a starter profile with an LLM

Paste a prompt like this into any chat model, filling in the researcher’s name and ORCID:

Generate a profile.jsonld file for the researcher-profiles format
(https://profiles.databio.org/context/v1.jsonld). The researcher is
<Full Name>, ORCID <0000-0000-0000-0000>.

Return a single JSON object with these required keys:
- "@context": "https://profiles.databio.org/context/v1.jsonld"
- "@id": "https://orcid.org/<their ORCID>"
- "@type": "Person"
- "conformsTo": "https://profiles.databio.org/context/v1.jsonld"
- "name": their full name
- "rid": their ORCID (no URL, just the identifier)
- "provenance": "third_party"
- "level": "lite"

Also fill in whatever of the following you can determine from public
information: "affiliation", "field", "subfields" (array), "summary"
(one paragraph), "expertise" (array of short topic labels).

Output only the JSON object, no commentary.

Example output:

{
  "@context": "https://profiles.databio.org/context/v1.jsonld",
  "@id": "https://orcid.org/0000-0002-1825-0097",
  "@type": "Person",
  "conformsTo": "https://profiles.databio.org/context/v1.jsonld",
  "name": "Jane Doe",
  "rid": "0000-0002-1825-0097",
  "provenance": "third_party",
  "level": "lite",
  "affiliation": "Example University",
  "field": "Computational Biology",
  "subfields": ["epigenomics", "chromatin"],
  "summary": "Jane Doe develops methods for region-set analysis.",
  "expertise": ["region-set-analysis", "reproducible-workflows"]
}

Save this as <slug>/profile.jsonld. LLM-generated metadata should be reviewed for accuracy before you treat it as authoritative: the model may guess at affiliation, field, or summary details it has not verified.

Validate the result

Run the conformance check. It validates every artifact against its schema and checks the cross-file invariants (manifest drift, summary ids that match a paper, and so on):

rp validate jane-doe

The same check from Python returns a ProfileValidationReport:

from researcher_profiles import ResearcherProfile

report = ResearcherProfile.from_files("jane-doe").validate()
print(report.ok)

Loading with from_files alone is not a conformance check: it reads each file lazily and only parses the ones you touch.

See How to validate a profile for schema validation without installing the package, and for the deeper quality-audit checks.