OneGoodEmail

Find the right person at a company and write them one good, human-sounding outreach email — résumé in, recipient and draft out.

Features

• Recipient lookup — type a company name (or paste its site / job-posting URL) and get the founder/CEO's email. Anthropic web_search resolves the domain + person in one call; Hunter verifies the address.
• Confidence tiers — every result is labelled ✓ verified, ⚠ likely (confirm first), or 💡 suggested (verify). A miss never errors; it degrades to a firstname@domain suggestion plus guidance.
• Email drafting — upload a résumé + paste the JD (+ optional writing-style sample) to get a tailored draft that's edited to strip AI tells and dash punctuation.
• Stateless by design — nothing is written to a database; lookups are single, assisted, 1:1.

Architecture

Next.js (App Router, deploy on Vercel)
├── Frontend (app/page.tsx + components/)
│     • Upload resume + paste JD (+ optional style sample)
│     • Draft display + inline edit + copy
│
├── API route (app/api/generate/route.ts, Node runtime)
│     • Resume parsing — lib/resume-parser.ts (unpdf for PDF, mammoth for DOCX)
│     • LLM prompt chain — lib/pipeline.ts
│         ① Extract structured resume info      (prompts.extractResumePrompt)
│         ② Parse JD key requirements           (prompts.parseJobPrompt)   ┐ run in parallel with ①
│         ③ Match + draft email                 (prompts.draftEmailPrompt)
│         ④ AI-flavor self-check & rewrite       (prompts.deAiPrompt)
│     • Returns the finished email + pipeline insight
│
└── API route (app/api/find-recipient/route.ts, Node runtime)
      • Company name OR url → founder's email — lib/recipient-finder.ts
          ①+② Resolve domain + founder via Anthropic web_search   (prompts.findDomainAndFounderPrompt)
              (a pasted URL skips the domain search — domain parsed in JS)
          ③  Verify / discover the address via Hunter             (lib/hunter.ts)
          ④  Tier by confidence: verified / likely / guess — never errors
      • Returns a RecipientResult (email, confidence, founder, sources, note)

Key files

Path	Responsibility
app/page.tsx	Landing page shell
components/EmailGenerator.tsx	Upload/paste form, calls the API
components/RecipientFinder.tsx	Company → founder email lookup UI + confidence badge
components/ResultEditor.tsx	Editable draft + copy (incl. To: line) + pipeline insight
app/api/generate/route.ts	Multipart entry point; parses file, runs pipeline
app/api/find-recipient/route.ts	Company/url → founder email lookup
lib/resume-parser.ts	PDF/DOCX/TXT → plain text
lib/llm.ts	Anthropic client + JSON-completion helpers (incl. web_search variants)
lib/prompts.ts	The prompt-chain step builders
lib/pipeline.ts	Orchestrates the resume → email chain
lib/recipient-finder.ts	Orchestrates the company → founder-email lookup
lib/hunter.ts	Hunter Email Finder wrapper
types/index.ts	Shared types across UI + server

Model

Two models, set independently:

• Email writing (resume → draft → rewrite) uses Claude Sonnet 4.6 (claude-sonnet-4-6). Override with ANTHROPIC_WRITING_MODEL.
• Recipient finding (company → founder/email web search) uses Claude Opus 4.8 (claude-opus-4-8) for stronger research reasoning. Override with ANTHROPIC_RECIPIENT_MODEL.

Setup

pnpm install
cp .env.example .env.local   # add ANTHROPIC_API_KEY (required) + HUNTER_API_KEY (optional)
pnpm dev

HUNTER_API_KEY is optional: without it the recipient lookup still resolves the founder and domain, but skips verification and returns a suggested firstname@domain address instead of a Hunter-verified one.

Open http://localhost:3000.