How it works

Evidence grounding

Jobs, employers, dates, certifications, and accomplishments come from your resume, never invented. Every adapted bullet maps to a specific line in what you pasted, and the mapping is exposed in the Analysis tab so you can verify row by row.

When the JD asks for something your resume doesn't demonstrate, we flag the gap honestly rather than fill it in. That's the line we don't cross.

Hallucination detection

Two modes, set per-account from Account → Preferences:

• Balanced (default). Generated claims with no source backing are kept but tagged with a “may not be verified” badge so you can review + edit them before exporting. This is the right mode for most resumes because the cost of a false-positive (deleting a real but poorly-worded line) outweighs the cost of a flag the user can fix.
• Strict (paid plan). Unsupported claims are removed and the LLM is asked to regenerate using only verifiable facts. Best for federal applications, medical credentialing, and any context where a single fabricated line could disqualify the candidate.

And because every adapted line is grounded in your resume, the draft shows a “from your resume” marker on each one you can click to read the exact source line it was built from. If a line can’t be traced with confidence, no marker is shown we never point you at a source we can’t prove.

Anti-fabrication validator

Every LLM response passes through a Pydantic-schema validator before it reaches your screen. Outputs that fail validation (missing required fields, inventing job titles or dates, or claiming credentials we never gave it) are rejected and the model is asked to try again.

This sits inside the engine layer (not the UI), so the same guarantees apply whether you're running Quick Draft, Cover Letter, or Interview Prep.

Resume-class aware routing

Eleven resume classes are auto-detected (federal, medical, academic CV, corporate, sales, etc.). The classifier decides which prompt template runs. Federal resumes get the KSA-formatted output, academic CVs stay as-is, and corporate resumes get the bullet-heavy rewrite. Wrong class? Override on the Draft input page before generating; the override flows through to Cover Letter and Interview Prep so all three tools use the same source of truth.

Multi-LLM routing

We route through OpenRouter to the best model per task: Anthropic Claude for long-context reasoning, OpenAI for structured JSON, Google Gemini for cost-efficient bulk generation and as fallback. If one provider is degraded or rate-limited, the next in the chain takes over automatically. See Subprocessors in our privacy policy for the full list of LLM vendors + their data- handling postures.

Detailed Draft: your tone, your gaps

Quick Draft does the heavy lift; Detailed Draft gives you the precision tools. Inside Detailed Draft you can:

• Add what's missing. When the JD asks for a skill or accomplishment your resume didn't demonstrate, add it via the “Add missing” sidebar. We'll rewrite the surrounding paragraph to integrate the new line in your voice, not the LLM's default tone.
• Tone-of-voice rewrite. Paste a 2-3 sentence sample of how YOU write (e.g. an old cover letter, a LinkedIn-About paragraph) and the engine adapts subsequent rewrites to match cadence, formality level, and vocabulary. The sample stays session-local, never persisted to the LLM's training feedback channels.
• Per-section regenerate. Don't love a single bullet? Right-click → Regenerate this section. The rest of the draft stays untouched.

Export & privacy

One-click PDF / DOCX export with the candidate's name baked into the filename (FirstName_LastName-AskTheRecruiter.docx). Sensitive PII (SSN, DOB, full home address) is auto- redacted before the LLM ever sees the text. See our privacy policy for the full pipeline and your account for the self-serve data export + deletion controls.