One designer-builder, a three-sided SaaS, shipped end to end on Next.js.

External proxies as discipline: competitive teardown, JTBD interviews with real dev directors from week one, and public IRS data as the trust-earning signal.

One Next.js stack, four properties

Marketing site, user app, admin portal, and enterprise surface in one Next.js codebase with shared design system, data model, and pipeline.

Marketing site: twenty-plus pages built to convert

Home shipped against an 11-beat B2B sales arc, pricing rebuilt after a HeyCatch audit, six audience and comparison pages, enterprise contact form with pre-filled audience hints, Strapi-powered blog. SEO per route with structured data.

User app: mission-first discovery for three audiences

Dashboard lands on mission-matched grants, with Browse (Funders, Funding, Nonprofits), My Radar, and Peer Intelligence sharing one component library. Agency Clients, Funder Grantees, and Funder Contacts (CRM-lite) ride the same shell.

Browse without noise

Browse Funders, Browse Funding, Browse Nonprofits all share one table pattern, one filter language, and one card style, so switching surfaces costs no learning.

The 990 is the thesis

Each funder surfaces typical award, award count, annual giving, focus areas, and recent recipients. The 990-driven view is what makes "mission-first" real.

Peer funders: who's backing orgs like mine

Peer Intelligence surfaces funders backing orgs whose work overlaps with the user's, generated by generate-peer-recommendations and refreshed as new 990 data lands.

Narrative angles: application strategy, generated

The system reads the org's programs, cross-references the funder landscape, and produces strategic framings with example funders and language the dev director can walk into an application with.

Admin portal: a one-person operator surface

Review Queue, News Queue, Content Batches, Funder Enrichment, Funders, Grants, Users, and Coverage, designed so one operator supervises a seventy-task pipeline without drowning.

Agentic ingestion pipeline: seventy-plus Trigger.dev tasks

• Discovery. sweep-archetypes orchestrates per-archetype sweeps. sweep-single-archetype and sweep-single-user search both the existing corpus and the open web via lib/ai/web-search.ts.
• IRS and ProPublica enrichment. bmf-import, ingest-funders-from-bmf, irs-filing-index-ingest, parse-990-pf, parse-990-schedule-i, parse-990-schedule-i-batch, funder-propublica-enrich, nonprofit-propublica-enrich.
• Funder enrichment from the open web. discover-funder-portal, discover-funder-portals-all, enrich-funder-from-website, enrich-funders-from-websites-all, suggest-funder-website, tag-funders-from-description, infer-funder-cycle.
• Matching. precompute-match-scores, sweep-paying-users, sweep-user-programs, refresh-archetype-coverage. Scoring runs ahead of time so the dashboard is instant.
• Peer recommendations. generate-peer-recommendations surfaces "funders backing orgs like yours."

Marketing automation: an end-to-end growth engine, no humans in the loop

Marketing on Funding Findr is not a side system. It is built into the same Trigger.dev + Resend + Buffer + PostHog stack as the product. One person can run it because it runs itself.

• Content generation and syndication. generate-content-batches drafts posts from the latest ingested grants and news. generate-batch-copy writes the social and blog copy. render-batch-carousels and render-batch-reel produce the social assets. publish-batch-blog ships the post. publish-batch-to-buffer queues it on social.
• News ingestion as a content source. poll-news-sources + extract-news-article pull philanthropy news. The admin News Queue approves it, then the content pipeline turns it into blog + social.
• Lifecycle email on Resend. send-welcome-email on signup. sync-user-to-resend keeps the audience list live. scan-abandoned-signups, trial-expired-sweep, card-expiry-sweep recover revenue. newsletter-nurture-sweep + weekly-briefing-sweep keep paying users engaged with their own grant matches, personalized per org.
• SEO as code. Six competitor and alternative pages (Candid, Instrumentl, GrantWatch, GrantStation, Hello Alice, Scholarships.com), audience pages for nonprofits and BIPOC founders, monthly deadline roundups auto-published as Trigger.dev batches, every page with structured data and JSON-LD.
• Analytics loop. PostHog captures every signup funnel event, which routes back into which content batches get queued next. Sentry catches the exceptions before a user does.

The whole system was built so one person could run a marketing engine that would normally require a growth team.

Multi-model AI via OpenRouter

One gateway, model-per-task routing: reasoning model for matching, cheap models for classification, long-context for narrative. Swapping a provider is a config change.

Design system: Luxury Minimal

Source Serif 4 for headlines, geometric sans for UI, warm monochrome palette with a single indigo accent on the subscribe button. Tokenized in Tailwind so a new comparison page takes a day.

Be every role, pick one at a time

Solo discipline is not multitasking, it is picking which role is in the chair this week. Research weeks run independent of design weeks and build weeks.

Sequence by business risk, not scope

Ingestion shipped before UI polish because discovery quality is the thesis. Comparison pages shipped before enterprise because positioning against Candid and Instrumentl is what converts.

Build guards so one person cannot forget

Every price string linted against lib/pricing.ts by a prebuild guard. Every ingested item passes a review queue before publish. Every long-running task has retries, observability, and alerting.