Production2023-2024E-Commerce + Computer VisionSolo Full-Stack Developer & ML Engineer

Buy It - AI-Powered TV Fashion Discovery

Transforming TV Viewing Into Instant Shopping Experiences

Solo-built AI-powered platform that automatically identifies and curates shoppable fashion products from TV shows. Users discover and shop outfits worn by their favorite characters through end-to-end system: Next.js frontend, Python/FastAPI backend, computer vision pipelines (YOLO), and cloud infrastructure on GCP.

10,000+

Products Indexed

Fashion items automatically detected and matched to e-commerce inventory across 500+ TV shows

500+

TV Shows Processed

Complete seasons from popular streaming platforms (Netflix, HBO, Hulu, Prime Video)

100,000+

Frames Analyzed

Video frames processed through YOLO detection pipeline with scene detection optimization

92%

Detection Accuracy

YOLO fashion detection accuracy on test set of clearly visible fashion items

85%

Product Match Rate

Successfully matched detected fashion items to relevant e-commerce products via affiliate APIs

The Challenge

TV viewers frequently want to buy fashion items they see on screen, but identifying and finding these products is extremely difficult. Current solutions require manual curation (expensive, doesn't scale), reverse image search (unreliable for video frames), or browsing generic affiliate links (poor conversion). The e-commerce opportunity is massive, but no one had solved the technical challenges of automated product detection, identification, and matching at scale.

❌

Manual fashion curation from TV shows is expensive and doesn't scale beyond a few popular series

❌

Viewers screenshot fashion items but have no way to identify and purchase them

❌

Existing 'shop the look' services cover <1% of TV content and require human editors

❌

Reverse image search fails on video frames due to motion blur, lighting, and angles

❌

Fashion affiliate marketing limited by inability to match on-screen items to e-commerce inventory

❌

No automated solution exists for large-scale video fashion extraction and product matching

The Solution

Built a complete AI-powered pipeline that automatically processes TV show videos, detects fashion products using computer vision, identifies and matches them to e-commerce inventory using multi-modal AI, and delivers results through a fast, scalable web application. Entire system built solo: frontend (Next.js/TypeScript), backend (Python/FastAPI), ML pipelines (YOLO + Gemini Vision), and cloud infrastructure (GCP + Terraform).

Tech Stack

Next.jsTypeScriptReactPythonFastAPIPyTorchYOLOGemini VisionFirestoreRedisDockerGCP Cloud RunTerraform

System Architecture

Frontend (Next.js)

Next.js 14 App Router, TypeScript, React, Tailwind CSS

Server Components for SEO, Client Components for interactivity, Virtual scrolling for 10K+ products, React Query for state management, Optimistic updates, Image optimization with next/image

Backend API (FastAPI)

Python 3.11, FastAPI, Pydantic, Firestore, Redis

Type-safe API with Pydantic models, Async/await for high concurrency, Redis caching layer, Firestore for product/show data, Background task queues, Rate limiting and auth middleware

ML Pipeline (YOLO + Gemini)

PyTorch, YOLOv8, Ultralytics, Gemini Vision API, OpenCV

Custom-trained YOLO model (92% accuracy), Scene detection for frame extraction, Gemini Vision for product identification, Batch processing with rate limiting, Quality scoring and filtering

Product Matching

Amazon PA-API, SERP API, Google Shopping, CLIP, sentence-transformers

Multi-source affiliate search with fallbacks, Semantic similarity with sentence-transformers, Image similarity with CLIP embeddings, Intelligent ranking algorithm, Deduplication via fuzzy matching

Cloud Infrastructure (GCP)

Cloud Run, Firestore, Cloud Storage, Memorystore (Redis), Terraform

Containerized microservices, Terraform IaC, Autoscaling 0-100 instances, Multi-stage Docker builds, Cloud Logging and Monitoring, IAM and security best practices

DevOps & CI/CD

GitHub Actions, Docker, Terraform, Cloud Build

Automated testing and type checking, Security scanning with Trivy, Staged deployments (dev/staging/prod), Zero-downtime deployments, Infrastructure as code with Terraform

Want to build AI-powered computer vision applications for e-commerce?

I specialize in end-to-end ML product development: from training custom detection models to building scalable cloud infrastructure. Let's discuss how to bring AI to your product.

Schedule a Consultation