Architecting and building a full B2B cold outreach automation SaaS — from database to AI pipeline to payment processing.

Drafter is a B2B sales outreach automation platform I designed, engineered, and shipped end-to-end. It helps freelancers and agencies automate their entire outreach workflow: researching target companies, finding decision makers via Apollo enrichment, and generating personalized cold emails, LinkedIn DMs, and Upwork proposals using AI. Every lead flows through a 10-stage sales pipeline — from discovery to meeting booked — with full tracking, analytics, and billing. Built with Next.js 15 (App Router), TypeScript, Supabase (Postgres + Auth + Row Level Security), Stripe for billing, Claude and Gemini for AI generation, and n8n for workflow automation. Deployed on Vercel with n8n self-hosted on Railway.

Building a production SaaS that orchestrates AI research, multi-provider enrichment, and automated outreach — all within a 10-second serverless timeout.

The core challenge was architectural: the AI research pipeline makes 4-5 LLM calls per lead (company research, decision maker research, email generation, LinkedIn DM generation, proposal writing) which takes 30-60 seconds total — but Vercel's Hobby plan has a 10-second function timeout. I had to design a split architecture where Vercel handles fast database reads/writes and n8n handles all the slow AI orchestration. On top of that, the platform needed to handle multiple discovery sources (job boards, LinkedIn, funded startups, signals), deduplicate leads across all sources, manage encrypted API credentials per user, enforce usage-based billing limits, and produce consistent AI output quality across different generation paths.

A complete outreach automation platform — 28 API endpoints, 12 database tables, 4 n8n workflows, and a full sales pipeline.

Drafter shipped with a comprehensive feature set: auth flow with onboarding, profile management with brand voice and writing samples, 7 document types for generation, a 10-state opportunity pipeline with keyboard shortcuts and bulk actions, Upwork job finder with RSS scraping and AI relevance scoring, signal hunting for funding rounds and hiring activity, autopilot sourcing across job boards and LinkedIn, analytics with conversion funnels, Stripe billing with Free/Pro tiers and usage limits, Gmail integration for direct sending, Apollo integration for decision maker enrichment, and Sentry for error monitoring. The domain-driven architecture keeps all of this maintainable.

Domain-driven architecture with split serverless + n8n orchestration, multi-provider AI, and production-grade infrastructure.

Designed the system around a domain-driven folder structure: features/ for UI components, domain/ for type definitions and interfaces, infrastructure/ for Supabase and API implementations, and shared/ for stores, layout, and common components. Built 28 API endpoints across research, generation, opportunities, signals, Upwork jobs, billing, and integrations. The research pipeline splits between a fast cache-check on Vercel and full AI orchestration on n8n — with intelligent caching (7-day TTL), company name deduplication, and automatic opportunity status progression as each stage completes.

Built 2 production n8n workflows (Research Batch and Autopilot Sourcing) with infrastructure ready for 2 more (Signal Hunting and Upwork Finder). The Research Batch workflow handles conditional branching between cached vs new research, basic vs deep research paths, and multi-step AI generation — all with rate limit awareness and retry logic. Implemented Stripe billing with webhook-driven counter resets, encrypted credential storage in Supabase's private schema with RLS, Upstash Redis rate limiting, Zustand for state management with optimistic updates, and a fully responsive mobile experience with a grouped bottom nav system.