FAQ
Frequently asked questions
Answers to the questions we hear most. If something isn’t covered, the resources page has a contact link.
Is my data sent to the cloud?
No. The LLM that writes your summary runs locally on your machine through Ollama — prompts never leave the host. PDFs are processed in memory and discarded as soon as the request returns. Results live only in your browser’s session storage. ClinVar lookups are the one outbound network call (to NCBI), and they only contain gene names and variant identifiers, never anything from your file. Read the data lifecycle →
Which lab reports are supported?
Lab-specific parsers currently cover Invitae cancer panels (positive and negative), Invitae carrier-screening report headers, and GeneDx whole-exome (XomeDx) reports. Other layouts are detected as “unknown” — in that case the parser returns a low- confidence signal and routes you to manual entry rather than guess.
How accurate is the variant extraction?
Extraction is regex- and layout-based, not ML, so it is deterministic. We run an accuracy harness against a small public corpus on every change and gate at ≥95% field accuracy. If a variant looks wrong on the results page, do not act on it — bring the original PDF to your counselor.
Should I make medical decisions based on the summary?
No. GeneTranslate is an educational aid. It deliberately avoids diagnostic language, treatment recommendations, and quantitative risk for high-risk genes. Every clinical decision should go through a board-certified genetic counselor or physician.
What if my report isn’t recognized?
If automatic parsing fails or returns low confidence, GeneTranslate prompts you to use manual entry: type the gene, HGVS notation, classification, condition, and zygosity for each variant. The downstream LLM analysis works the same way.
Why are some risk numbers hidden?
Findings in high-risk genes such as BRCA1, BRCA2, the Lynch genes (MLH1, MSH2, MSH6, PMS2, EPCAM), TP53, PALB2, ATM, CHEK2, STK11, CDH1, and MUTYH trigger an elevated safety mode. Quantitative lifetime risk estimates are removed and a banner routes you back to a counselor. These are conversations a clinician should have with you, not a tool. More on the safety layer →
Can I export or share my results?
Yes. From the results page you can export a PDF (jsPDF) or email a copy through Resend. Email addresses are passed through to the email provider for delivery and are not retained server-side or client-side after the send completes.
Do you store anything about me?
No accounts, no analytics, no cookies, no databases. The product is intentionally stateless. If you close the browser tab, your results are gone — export the PDF or email it to yourself if you want a copy.
Which LLM does it use?
The default is
qwen2.5:7b-instruct running through Ollama. Any JSON-capable Ollama model can be swapped in via configuration. We use JSON mode to keep responses structured and parseable.