Scene 1 (0s)
Behind The Search Bar. How google search work The engineering behind instant results How does a short query return useful results in under a second?.
Scene 2 (12s)
Coming up…. 01 End-to-end pipeline: Crawl → Index → Retrieve → Rank → Serve.
Scene 3 (13s)
Coming up…. 01 End-to-end pipeline: Crawl → Index → Retrieve → Rank → Serve.
Scene 4 (31s)
SEARCH PIPELINE. five core stages. SERVE Deliver the final ranked results instantly, adding personalization and safety filters to enhance the user’s experience. 🌐.
Scene 5 (1m 1s)
SERVE Deliver the final ranked results instantly, adding personalization and safety filters to enhance the user’s experience. 🌐.
Scene 6 (1m 34s)
[image] Crawling & Indexing Googlebot-style crawler respects robots.txt, prioritizes seeds, discovers links, updates pages incrementally. Index stores mappings from terms to document (inverted index). Key features: tokenization, metadata, signals like freshness and PageRank..
Scene 7 (2m 5s)
Googlebot-style crawler respects robots.txt, prioritizes seeds, discovers links, updates pages incrementally. Index stores mappings from terms to document (inverted index). Key features: tokenization, metadata, signals like freshness and PageRank..
Scene 8 (2m 20s)
Maps terms ➡️ list of documents containing them. Each posting may include frequency and position info. Designed for speed with compression and skipping techniques..
Scene 9 (2m 33s)
Candidate generation: find documents matching query terms. Scoring uses models like TF-IDF or BM25 to rank relevance. Modern search adds machine learning rerankers on top..
Scene 10 (2m 45s)
Combines multiple signals: lexical relevance, PageRank, freshness, and user context. Learning-to-Rank models reorder candidates for better relevance. Neural methods (e.g., embeddings, BERT rerankers) capture semantic meaning..
Scene 11 (3m 0s)
Infrastructure – Speed & Scale Sharding & replication: distribute index across many servers. Caching: store frequent queries and results. Approximate nearest neighbour (ANN): accelerates vector-based search. Coordinators merge results from multiple shards in milliseconds..
Scene 12 (3m 23s)
Infrastructure – Speed & Scale Sharding & replication: distribute index across many servers. Caching: store frequent queries and results. Approximate nearest neighbour (ANN): accelerates vector-based search. Coordinators merge results from multiple shards in milliseconds..
Scene 13 (3m 31s)
Key Takeaways. summary of insights. 1. 2. 3.
Scene 14 (3m 31s)
A simple query sets off a complex process: Google finds and delivers results by crawling web pages, storing them in an index, and ranking them based on relevance and quality. How search works.
Scene 15 (4m 3s)
A simple query sets off a complex process: Google finds and delivers results by crawling web pages, storing them in an index, and ranking them based on relevance and quality. How search works.
Scene 16 (4m 33s)
Happy to take your questions!. Presented by: Hardeep Singh Bohra.