Prepare
Remove names, dates, record numbers, addresses, and other direct identifiers. Use a filtered genomic excerpt.
A TRANSPARENT RESEARCH PIPELINE
From raw signal to
reviewable evidence.
RDK does not preload a diagnosis or pretend to know the answer. It waits for an authorized researcher to provide de-identified inputs, then displays only what the live OpenAI analysis returns.
Open the workspace โ
Describe once
Type one de-identified case description. A VCF or phenotype file is optional and can strengthen the evidence.
Analyze
The server sends the authorized excerpts to OpenAI, requests structured output, and searches current scientific sources.
Review
Challenge every ranking, rationale, source, limitation, and validation requirement before research use.
What appears after a live run
Variant rankings
When genomic data is supplied: gene, variant notation, predicted consequence, model-assigned evidence score, confidence label, and evidence. Without genomic data, RDK leaves this section unranked rather than inventing variants.
Evidence synthesis
A concise case summary, a mechanism explanation, and a visual gene-to-mechanism map generated from the same response.
Research hypotheses
Each direction includes its title, mechanistic rationale, evidence level, and the exact experimental or clinical validation the model says is still required.
Provenance
Clickable sources, source relevance, limitations, model name, response identifier, token usage, and the model's own research disclaimer.
No silent sample swap. Empty means no analysis has been run in that browser. A populated result means the interface received a successful live API response.