M0
Available
Free
The Problem
Ranking audit
You can articulate why relevance is hard and what signals a ranker needs
IndexZero
Build a search engine from scratch in Python
Ten modules. One Python codebase that grows. You start with ranking judgment and end with a working search API. Along the way, you build every piece yourself: tokenizer, inverted index, BM25 scorer, eval harness, vector retrieval, and hybrid search.
Build arc
One codebase, one chain of system states
Each module changes the system state. The next module starts from the artifact you just made.
- M0The ProblemRaw titles become ranking hypotheses
- M1Text ProcessingText becomes tokens
- M2The IndexTokens become postings
- M3RankingPostings become ranked results
- M4Did It Work?Rankings become metrics
- M5Smarter QueriesQueries become structured
- M6Meaning, Not WordsMeaning becomes vectors
- M7Both TogetherSignals become one ranked list
- M8Keeping It AliveIndexes become live systems
- M9The Full SystemSystem becomes an API
Modules
Ten modules from ranking judgment to search API
Free modules let you start the system. The cohort continues through indexing, ranking, evaluation, query processing, and deployment shape.
M1
Available
Free
Text Processing
Tokenizer + vocabulary
A working tokenizer that turns raw text into clean token streams
M2
Available
Cohort
The Index
Inverted index + lookup
A searchable inverted index that maps terms to documents
M3
Available
Cohort
Ranking
BM25 scorer
A BM25 scorer that ranks documents by relevance to a query
M4
Planned
Cohort
Did It Work?
Eval harness
An evaluation harness that measures ranking quality with labeled queries
M5
Planned
Cohort
Smarter Queries
Query processor
A query processor that handles multi-word and structured queries
M6
Planned
Cohort
Meaning, Not Words
Vector retrieval
Vector embeddings and approximate nearest neighbor search over the same corpus
M7
Planned
Cohort
Both Together
Hybrid retrieval
A hybrid retriever that combines lexical and semantic signals
M8
Planned
Cohort
Keeping It Alive
Index pipeline
An index that handles document changes without full rebuilds
M9
Planned
Cohort
The Full System
Search API
A complete search API serving queries over your index
Start free
M0 and M1
Start with M0 and M1. You define the ranking problem, build the tokenizer, and see how the corpus changes before indexing starts.
- Ranking audit
- Tokenizer and vocabulary
- Tests and assessments included in the repo
Join the cohort
M2 and beyond
Continue with M2 and beyond through the guided cohort. You get code reviews, deadline structure, discussion, and a workshop after the ranking modules.
- Inverted index and ranking modules
- Code reviews and deadline structure
- Workshop after the ranking modules
Evidence
Concrete workload, one dataset, compounding artifacts
One codebase across 10 modules. M1: 44 tests, 7 assessments. M2: 26 tests, 6 assessments. M3: 27 tests, 5 assessments. Real product dataset. Each module feeds the next.
Who built this
About the instructor
IndexZero is built by Sumit Garg. Sumit spent years building search infrastructure at Microsoft, working on the systems behind Azure AI Search that handle billions of queries. He built this course because most developers use hosted search without understanding the retrieval mechanics underneath. IndexZero makes those mechanics visible in code you write yourself.