Donna
An AI-assisted job search copilot, from finding roles to landing the offer. Donna drafts and recommends; you review and submit everything. It never auto-applies, never sends email, and never scrapes sites that forbid it.
Meet Donna
- Voice-led onboarding. Donna greets you, asks your name, connects your AI key, then reads your CV (PDF, DOCX or text), a LinkedIn PDF export and your website, and builds your profile. Her voice is a natural-sounding open-source model (Kokoro), pre-recorded and shipped with the app, so it is free for every user. Optional premium (OpenAI) or custom voice servers are in Settings. See
docs/VOICE.md.
- A real profile, not a text blob. Roles, achievements, projects, education and skills are separate components you edit in place, laid out like read.cv. Preview it, or export it as a standalone HTML page. A shareable public link is planned (the
handle field is reserved).
What it does
| Stage | Feature | Borrowed from |
|---|
| Find | Paste link/text, official Greenhouse/Lever/Ashby/RemoteOK feeds, job-alert email import, one-click bookmarklet | career-ops (portals), ai-job-search (scrape + dedupe) |
| Fit | 1-5 score across 5 dimensions, deal-breaker vetoes, legitimacy observations, gaps | career-ops (A-H eval), ai-job-search (dimensions, vetoes) |
| Tailor | CV + cover letter from your master CV only, separate reviewer pass that fact-checks, unsupported keywords listed as gaps | ai-job-search (drafter/reviewer), career-ops (never fabricate) |
| Export | PDF (default), Word, Markdown; 3 templates; switch any time after generation | - |
| Apply | Answer bank + AI drafted answers to form questions; you click submit | Auto_job_applier (answer bank, dry-run spirit) |
| Track | Board/table pipeline, statuses, follow-up reminders | career-ops, ai-job-search |
| Interview | Prep pack, STAR+R story bank, recruiter/follow-up/thank-you drafts | career-ops, ai-job-search |
Bring your own key: Anthropic, OpenAI, or any OpenAI-compatible endpoint (OpenRouter, Groq, ...). All AI calls go through one interface (src/lib/ai.ts), so a hosted "full service" tier can be added later as another provider.
Run locally
npm install
npm run dev # http://localhost:3000
npm test
Then: Settings (add key) → Profile (paste CV) → Add jobs.
Deploy free on Vercel
- Push this repo to GitHub (private is fine).
- On vercel.com: Add New → Project → import the repo → Deploy. No environment variables needed.
- Your app is live at
https://<project>.vercel.app.
Note: the Hobby plan is for non-commercial use. Move to Pro when this becomes a paid product.
Data and privacy (v1)
Data is stored in your browser's localStorage (back it up from Settings). Your API key stays in the browser and is relayed per request by /api/ai, which does not store or log it. Persistence is isolated in src/lib/store.tsx; supabase/schema.sql is the schema (with row level security) for the planned move to Supabase.
Safety choices
- LinkedIn/Indeed/Glassdoor are never fetched server-side; paste or use the bookmarklet in your own session.
- Server fetches block private/internal addresses (SSRF guard).
- Faithfulness rules in every prompt: reformulate, never invent.
Roadmap
-
Public shareable profile at a handle (needs accounts)
-
Supabase auth + sync (schema ready)
-
Hosted "full service" AI tier and billing
-
Gmail/Notion sync, reply classification, funnel analytics (career-ops, ai-job-search)
-
Browser extension for form-fill assist (user clicks submit)
-
Import your Claude project's prompts and rules (src/lib/prompts.ts)
