Merchant Onboarding & Enrichment (Firebase + Genkit AI)
A crypto payments super-app
Overview
A build exploring AI-assisted merchant onboarding and enrichment for the payments super-app, built on Firebase with Google’s Genkit framework. It combines web crawling, LLM processing and background jobs to turn raw merchant inputs into structured, enriched records.
Why It Exists
Onboarding merchants at scale means gathering and normalizing scattered web information. This build tested whether an LLM pipeline could automate merchant data enrichment and reduce manual onboarding effort.
What We Built
A Firebase project (firebase.json, Firestore rules and indexes) whose Cloud Functions (functions/) orchestrate an AI pipeline using Google Genkit with the genkitx-openai plugin and the Genkit evaluator for output quality, Firecrawl (@mendable/firecrawl-js) and url-metadata for web crawling/metadata extraction, and Trigger.dev (@trigger.dev/sdk) for durable background jobs. firebase-admin and Firestore provide persistence.
Technologies & Approach
Genkit was chosen for a structured, evaluable LLM workflow on Firebase; Firecrawl/url-metadata for content acquisition; Trigger.dev to run long-running enrichment off the request path. Firebase Functions + Firestore enabled a fast, serverless build with no infra to manage.
Outcome / Impact
Validated an end-to-end AI enrichment flow, crawl, extract, LLM-process, evaluate, persist, for merchant onboarding, demonstrating feasibility before any production investment.
Capabilities Demonstrated
- LLM pipelines for automated data enrichment (Google Genkit + OpenAI)
- Web crawling/metadata extraction integration (Firecrawl)
- Serverless, durable background processing (Firebase Functions + Trigger.dev)
- Rapid, evaluable AI build delivery