I'm building personalized homepage recommendations, what stack should I use?

A good stack depends on your scale, latency needs, and how “personalized” you want the homepage to be. For most teams, I’d recommend a hybrid stack that separates event collection, feature storage, model serving, and experimentation.

A solid practical stack

1) Frontend / homepage delivery

• Next.js or React for the homepage
• Server-side rendering or edge rendering if you need fast personalized page loads
• CDN: Cloudflare, Fastly, or Vercel Edge

2) Event tracking

You need user behavior signals like:

• page views
• clicks
• dwell time
• searches
• purchases / conversions

Use:

• Segment, RudderStack, or a custom event pipeline
• Send events into a streaming system like Kafka, Kinesis, or Pub/Sub

3) Data storage

Use two layers:

• Warehouse for analytics/training: BigQuery, Snowflake, or Redshift
• Operational store for low-latency user features: Redis, DynamoDB, or Postgres depending on scale

4) Feature store

If you’ll do ML-based personalization:

• Feast is a strong open-source choice
• Or managed options if you want less ops overhead

Store:

• user embedding
• recent clicks
• session recency
• category affinities
• content popularity trends

5) Recommendation engine

Pick based on maturity:

If you want fast startup and strong baseline:

• Content-based recommendations
• Popularity + recency + personalization rules
• Good for MVPs and cold start

If you want ML recommendations:

• Python stack with PyTorch or TensorFlow
• Models:
  • collaborative filtering
  • learning-to-rank
  • two-tower retrieval
  • gradient-boosted ranking with XGBoost/LightGBM

A common architecture:

• Candidate generation: retrieve ~100–1000 items
• Ranking model: score and order the final homepage items

6) Model serving

• FastAPI or gRPC service for real-time inference
• TorchServe, BentoML, KServe, or custom serving
• Cache top results in Redis for speed

7) Experimentation / A-B testing

Essential for recommendations:

• GrowthBook, Statsig, Optimizely, or in-house
• Track CTR, conversion, retention, dwell time, revenue

8) Monitoring

• Prometheus + Grafana for infra
• Evidently AI or custom dashboards for drift / model quality
• Monitor latency, error rates, CTR, and business metrics

---

Recommended stack by company stage

MVP / small team

Use:

• Next.js
• Postgres
• Redis
• BigQuery or Postgres analytics
• Python + LightGBM
• FastAPI
• GrowthBook

This is simple, cheap, and works well for rule-based + lightweight ML recommendations.

Growth stage

Use:

• Next.js
• Kafka + warehouse
• Feast
• PyTorch/LightGBM
• FastAPI/KServe
• Redis cache
• Statsig/GrowthBook

Better for scale, freshness, and experimentation.

Large scale / high personalization

Use:

• Event streaming: Kafka/Kinesis
• Feature store: Feast + online store
• Retrieval model: two-tower embeddings
• Ranking model: LTR / deep model
• Search/vector DB: Milvus, Pinecone, Weaviate, or pgvector
• Real-time inference with caching and fallbacks

---

If you want my opinionated default

If you’re starting now and want a good balance of simplicity and power:

• Frontend: Next.js
• Backend API: FastAPI
• DB: Postgres
• Cache: Redis
• Analytics warehouse: BigQuery
• ML: Python + LightGBM to start, PyTorch later
• Experimentation: GrowthBook
• Infra: Docker + Kubernetes only if needed
• Tracking: Segment or RudderStack

This stack is practical, avoids overengineering, and can grow with you.

If you tell me:

1. your team size,
2. expected traffic,
3. whether this is content, ecommerce, or social,
4. and whether you need real-time personalization,

I can recommend a much more specific stack and architecture.