Scooper — AI-Powered Content Intelligence Platform

Designed and built Scooper end-to-end: an AI-powered platform that aggregates newsletters, RSS feeds, and YouTube into personalized intelligence digests.

Unlike single-email summarizers, Scooper detects cross-source trends, extracts time-sensitive action items, and scores content relevance (0-100) based on your role and goals.

Key technical achievements:

AI cost optimization: 50% savings via Anthropic Batch API + smart filtering

Three-layer personalization: Explicit profile + behavioral learning + conversational refinement

Multi-format delivery: Email, audio (TTS), and web dashboard

Real-time processing visibility: Users see exactly what's being analyzed

Stack: Next.js 15, React 19, Convex, Anthropic Claude, Expo (mobile), Polar (payments)

Role: Product Engineer (AI/Full-Stack) — Solo Builder

AI/ML Integration

Full-Stack Development

Product Engineering

Next.js / React

TypeScript

System Architecture

Anthropic Claude API

Real-Time Applications

SaaS / Subscription Products

Mobile Development (React Native)

Knowledge workers subscribe to 100+ newsletters but open only 12%. Information overload leads to missed opportunities and decision fatigue.

Existing solutions fall short:

Gmail summarizes single emails — missing cross-source connections

Read-later apps require manual curation — adding more work

Generic AI tools lack personalization — treating everyone the same

I needed to build something that aggregates multiple sources, understands individual context, and delivers actionable intelligence in under 10 minutes.

Scooper aggregates content from newsletters, RSS, and YouTube, then uses AI to extract what matters to YOU specifically.

Connect your sources (forward newsletters, add RSS feeds, link YouTube channels)

AI analyzes every piece against your role, industry, and goals

Receive a personalized digest via email, audio, or web — your choice

Multi-source intelligence — Trends that only emerge when you see content together

Relevance scoring — Every item rated 0-100 based on your context

Action item extraction — Only genuinely time-sensitive items, not generic advice

Conversational tuning — Chat with AI to adjust preferences: "more AI topics, less funding news"

Three personalization layers (explicit + implicit + conversational)

50% AI cost reduction via batch processing

Real-time processing visibility

Multi-engine TTS for audio digests

Founding member pricing with progressive discounts

Frontend: Next.js 15, React 19, Tailwind CSS 4, Radix UI

Backend: Convex (serverless), 33 TypeScript modules

AI: Anthropic Claude (Haiku for bulk, Sonnet for chat)

Mobile: Expo + React Native

Payments: Polar subscriptions

Database: 18 tables with composite indexes, content hashing for deduplication, soft deletes, token budget tracking

Anthropic Batch API for scheduled digests (50% savings)

Pre-AI filtering (promo detection, quality scoring)

Adaptive summary tiers (200-2000 tokens based on priority)

Per-user token budgets

Shipped production-ready AI SaaS from zero in ~2 weeks

50% reduction in AI costs via strategic optimization

3 delivery formats (email, audio, web)

Real-time processing with full transparency

Founding member monetization active

Mobile app in development (Expo/React Native)

AI cost optimization is table stakes — Built batching and filtering from day one

Personalization compounds — Three layers >> one layer

Transparency builds trust — Showing processing status reduced confusion

Audio unlocks new use cases — Commute-friendly digests = daily habit

Founding member pricing converts — Urgency beats discounts