100 users, all hand-built 1:1. Live at grapevines.ai.
Recruiters publicly admit roughly a third of postings are pipeline-building, not real. Internal referrals are eighteen times more likely to land an interview than cold applications. Grapevines treats job search like B2B sales — positioning, qualifying, research, targeted outreach — and lets the resume be a downstream artefact, not the strategy. 100 users so far, every one onboarded through a one-on-one conversation.
Public lead magnet + the full app.
CV Intel at grapevines.ai/intel is the public diagnostic — drop a resume + a job, get back a read on who actually makes the call at the target company and what they care about. Never fabricates a person; if it can’t find the team, it says so and falls back to company-only intel. The full app sits behind it.
Senior job search is broken at the channel, not the resume.
Template-based ATS tools optimize for keyword score; they don’t move the needle on hiring-committee decisions. Human coaches at $300/hour deliver the strategic depth — positioning, target research, interview prep — but aren’t available for daily iteration. Grapevines is the wedge between the two: the coach’s frame, available at the cadence the job search actually runs at.
The customer-discovery work made the channel problem concrete. Roughly one in three job postings is pipeline-building. Some companies won’t schedule a first round without an internal recommendation. Industry baseline for cold applications is around 2% application-to-interview — and referrals are eighteen times more likely to land. The math says run a B2B sales motion, not a volume play.
Six surfaces, sequenced like a sales pipeline.
CV Intel · /intel
Public diagnostic + lead magnet. IntelOrchestrator chains Exa company research with EnrichLayer people search, identifies likely hiring managers + role-in-decision, never fabricates. Falls back to company-only if no team data.
Positioning Discovery
AI conversation that surfaces the candidate's positioning + bridge story — the narrative connecting where they've been to where they're going. Discovery proved this is the differentiator at senior levels, not skills.
Adaptive job rubric
Not a fixed 100-point rubric — a 6–8 category rubric generated per job, weighted by the user's top competencies against the role. If you've run 47 A/B tests, “experimentation” weighs 22pts for data-driven PM roles, not the generic 10.
Company research
Exa AI neural search → Perplexity / Claude fallback → team research agent → per-person assessment. Produces a buying-center map, not a careers-page summary.
Outreach identification
EnrichLayer + AI surfaces the right stakeholders with contact cards, connection angles, and conversation starters tied to the company-research findings.
Materials generation
Resume, cover letter, LinkedIn content — gated behind the research step. By the time materials get drafted, the system already knows the company, the role, and the people. They write themselves.
100 users. Every onboarding a one-on-one.
The growth strategy is the product strategy: every user came through a direct conversation. No paid ads, no growth hacks, no “sign up for the waitlist.” The reason is the work itself — positioning and target research only land when the system has heard the person in their own words first. That’s also what makes the system trustable enough to act on.
Discovery ran SVPG-format with nine one-on-ones (Ken Gordon, Priscilla Alves, Dheeraj Thakkar, Cory Hash, Josh, Jessica, Jon, Brett, Priscilla again on beta). The non-obvious learnings were the ones that re-shaped the roadmap: positioning as the actual differentiator (not output volume); scheduling + consistency moving visibility more than any single great post; human-in-the-loop non-negotiable.
Napa Dusk — career therapy with teeth.
The design system is named and locked: Napa Dusk. Lora serif for headings, Inter for body. Eight-color palette in fog / card / panel + ink + grape / vine / amber + border. 24px corner radius on cards — the signature rounded surface. Philosophy: therapeutic, not clinical. Story-first, not metrics-first.
The product reads more like a coach’s office than a SaaS dashboard. That’s deliberate. Senior career transitions are emotional; the system has to feel like it’s on the user’s side before they’ll trust it with their actual story.
Multi-agent, multi-vendor, with explicit fallbacks.
Frontend: Next.js 16 + React 19 + Tailwind on Vercel. Backend: FastAPI on Python 3.11, Docker on Render. Database: Supabase. AI: Anthropic Claude Sonnet 4 + Haiku as primary, OpenAI GPT-4o + GPT-4o-mini as secondary. Research: Exa AI for neural search, EnrichLayer for people data, Perplexity as a research fallback. Email: Resend. Analytics: PostHog + Mixpanel.
The fallback design is deliberate. Company research starts with Exa, drops to Perplexity if Exa misses, drops to Claude with structured prompting if both miss. People data is EnrichLayer first, inference second, “we couldn’t find them” third — the system would rather be honest than make someone up.