Systems Atlas

Chapter 0.1

Who This Guide Is For

Engineers, ML practitioners, product managers, and founders who want to go beyond "using a search API" to understanding how search systems actually work.


Primary Audiences

Software Engineers (Backend/Platform)

Profile: 2-5 years experience building APIs and services. Knows databases, REST, microservices. Has "used" Elasticsearch/Algolia but doesn't understand the internals.

📍 Common situation:

"I followed a tutorial to set up Elasticsearch. It worked for 10K products. Now we have 10M and everything is slow. I don't know where to start."

What you'll learn:

  • Move from "consumer of search API" to "builder of search infrastructure"
  • Understand trade-offs: Why does ranking matter more than retrieval?
  • Debug production issues: Why is P99 latency spiking?
  • Schema design, sharding strategies, and reindexing without downtime

ML Engineers / Data Scientists

Profile: Strong in embeddings, LLMs, recommendation systems. Weak in systems engineering (distributed systems, caching, latency).

📍 Common situation:

"My BERT reranker has 0.85 NDCG offline. But when we deployed it, CTR didn't change. The team says it's 'too slow' but I don't understand what that means for search."

What you'll learn:

  • How to take a model from Jupyter notebook to production search
  • The full pipeline: Retrieval → Ranking → Serving
  • Feature stores, model serving latency, where ML fits in
  • Why retrieval is the bottleneck (can't rank what you don't retrieve)
  • Training embeddings on click data, dealing with position bias

Product Managers (Technical)

Profile: Owns the search experience for an e-commerce or SaaS product. Reports to leadership on search KPIs.

📍 Common situation:

"I asked the team to 'add synonyms' and they said it would take 3 sprints. Why? Also, why can't we just use ChatGPT for search?"

What you'll learn:

  • Vocabulary to communicate with engineering: recall, precision, P99
  • Why some improvements are 2-week projects and others are 6-month investments
  • Framework for prioritizing: relevance vs latency vs personalization
  • How to read search dashboards and identify opportunities
  • When to push back on "it's too hard" vs when to trust the team

Founders / CTOs

Profile: Building a product where search is core (marketplace, knowledge base, etc.). Need to make build-vs-buy decisions.

📍 Common situation:

"We started with Algolia but it's costing $10K/month. Should we migrate to Elasticsearch? Also, our engineer says we need a 'vector database' now. What even is that?"

What you'll learn:

  • When to use Algolia vs Elasticsearch vs Typesense vs build custom
  • What's the minimum viable search stack for a startup?
  • How search affects retention and revenue (with numbers)
  • How to hire for search roles, what to look for
  • Red flags: over-engineering vs under-investing

Prerequisites

✓ You should know

  • • Basic programming (Python, JavaScript, or similar)
  • • What an API is and how HTTP works
  • • What a database is (SQL or NoSQL)
  • • Basic data structures (arrays, hash maps)

○ Nice to have

  • • Experience with Elasticsearch, Solr, or Algolia
  • • Basic understanding of distributed systems
  • • Familiarity with ML concepts (embeddings)
  • • Production experience with high-traffic systems

Who This Is NOT For

  • Complete beginners You need basic programming skills first. Try freeCodeCamp or similar.
  • Academic IR researchers This is practical, not theoretical. We skip the math proofs.
  • Copy-paste coders You won't find "paste this YAML" tutorials here. We focus on understanding.
  • People looking for quick fixes Search is complex. This guide respects that complexity.

Time Investment

2-3 hrs

To understand the fundamentals (Ch 0-3)

10-15 hrs

To complete the core curriculum (Ch 0-14)

30+ hrs

For deep mastery with exercises (All chapters)