LiveFounderDesigned, built, and operated solo

Marketplace Beta

77+
RSS sources ingested
01

The problem

Amazon, Walmart, and marketplace agency operators have no single credible resource tracking what is actually changing in their industry — and that audience is exactly the buyer profile worth reaching.

02

What I built

An e-commerce intelligence publication and aggregation platform, built end to end. A multi-model AI pipeline runs the editorial operation with model tier matched deliberately to task complexity against cost: Claude Haiku for classification, tagging, and triage where throughput and price dominate; Claude Sonnet for original editorial analysis where reasoning quality dominates; DALL·E 3 for imagery. Everything routes through the Vercel AI Gateway. Five scheduled stages handle fetch, classify, generate, illustrate, and cleanup, with a Resend-powered email layer for subscriber delivery. Embedded operator tools — PriceScope, AgencyForecast, and a recovery estimator — give the audience a reason to return.

03

What changed

Operates as an industry resource, an audience engine, and a lead-generation surface simultaneously. Presented internally at Threecolts as the company case study in applied AI.

Architecture

How it runs

VERCEL CRON — FIVE SCHEDULED STAGESSources77+ RSS feedsFetchingest · dedupeClassifyClaude HaikuGenerateClaude SonnetIllustrateDALL·E 3Cleanupprune · archiveVercel AI Gateway — every model call routed and meteredSupabase / PostgreSQLarticles · tags · subscribers · run logs · full-text search indexOUTPUTSmarketplacebeta.compublic publicationSubscriber emailResend · React EmailEmbedded operator toolsPriceScope · AgencyForecast · Recovery estimatorCyan marks a model call. Amber marks something live and reachable.

Stack

Next.jsTypeScriptTailwind CSSshadcn/uiSupabase / PostgreSQLVercelVercel CronVercel AI GatewayResendReact EmailClaude SonnetClaude HaikuDALL·E 3