Human, Actually

Most resume tools either invent experience you do not have or reshuffle the same words until the page reads like an AI wrote it. Human, Actually takes the opposite approach. It starts from everything you have actually done, your resume, your personal website, your GitHub, your writing samples, your notes, and builds a structured evidence base. Every claim about you gets traced back to a source document and labeled by strength: direct, inferred, or self-reported through an interview.

When you paste a job description, the system does not just keyword-match. It extracts the real requirements, then checks each one against your evidence. Binary screening requirements like work authorization, visa status, relocation, and clearance skip the LLM entirely and resolve from a simple Yes/No/Skip toggle, so the instant, deterministic answers never hallucinate. Everything else gets matched against your actual history, with gaps flagged for you to fill.

LinkedIn recommendations are treated as first-class third-party evidence. Import them from a CSV export and they corroborate the claims on your resume, show up as endorsement evidence on requirement cards, and inform how cover letters frame your strengths. This is the kind of supporting material most application tools ignore, even though it is often the most persuasive thing a candidate has.

When evidence is thin, the app runs an adaptive interview. It asks targeted questions to fill the specific gaps it detected, or you can chat freely and it extracts facts from the conversation. The output is a tailored resume, a cover letter that references the actual company and its values, interview prep stories, and on-demand answers to any application form question you paste in. You can download everything in Markdown or RTF.

Under the hood it runs on Next.js 15, Prisma with Postgres, OpenAI structured outputs with Zod validation, and a Prisma-backed job queue with a dedicated worker for long-running ingestion and analysis. Credentials for protected sites are encrypted at rest with AES-256-GCM or Google Cloud KMS and are never sent to the language model.