An AI-native photography platform — culling, captions, face-matching and a global directory — shipped end to end without an in-house AI team.

AI photo culling engine: Sharp + perceptual-hash de-duplication and scene grouping feed a Google GenAI vision pass that scores sharpness, exposure, composition and duplicates, sorting keepers from rejects with strict/balanced/lenient sensitivity — hours of culling collapsed into minutes

AI marketing ops engine: a multi-model pipeline (Claude Sonnet 4.6 + Haiku 4.5) that auto-selects portfolio shots and writes platform-aware captions, hashtags and a full social calendar, governed by per-platform hard rules, seasonal/platform/audience intelligence and a caption-scoring guardrail

Document-to-data directory pipeline: a Python scraper queries Google Places, then Claude Haiku 4.5 extracts structured photographer fields from messy websites, normalizes phones to E.164, dedupes, and bulk-upserts through the API as the single validation source

Productized media backbone: presigned S3 + FTP camera ingest, EventBridge-triggered Lambdas for thumbnailing/HLS transcoding and AWS Rekognition face indexing for selfie-to-photo matching, plus AI Reels and AI-designed QR event standees

SEO-grade public directory on Next.js 16: SSR profile/city/area/country pages with a sharded sitemap-index built to scale to millions of URLs, privacy projection so phone/email never render, and OTP claim-and-lead flows that turn listings into qualified enquiries

Cost and safety guardrails baked in: hard LLM spend caps and request belts per run, fingerprint-based resume to skip unchanged work, and a Playwright SEO/privacy matrix plus Lighthouse budgets (SEO 100, A11y 95+, Perf 85+) gating every release