How to Start a Career in Search
A self-directed learning path for breaking into search and information retrieval. Search is a specialized field — generic software or ML backgrounds get you partway, but the people who thrive learn the IR fundamentals, build an evaluation mindset, and follow the small, generous community that publishes openly.
This note is a curated entry point into the rest of the vault. Each step links to the deeper material.
The Short Plan
- Read the topics — start with practice-oriented guides, not theory
- Move to the concepts — learn the IR fundamentals that underpin everything
- Follow the people and companies — the field is small and publishes in the open
- Read the books — a handful of canonical texts, several free
- Read the case studies — learn how the industry leaders actually built it
- Take a course — even a free one builds structure
- Attend the conferences — practitioner events are where you meet the community
- Explore the tools — get your hands dirty with real engines
- Get active in the community — don’t just lurk; participate
- Monitor where the frontier is — track the vault’s year-by-year frontier index as the field moves
- Understand the roles — know which job you’re actually aiming for
The single trait every search hire is screened for is an evaluation mindset: the instinct to measure whether a change actually improved things. See Hiring for Search. Build this habit from day one and you are ahead of most candidates.
1. Read the Topics
The Topics folder is the fastest on-ramp — these are “how to do something in search” guides written for practitioners. A suggested reading order:
- Search Problem Archetypes — recognize the recurring shapes of search problems
- E-commerce Search — the most common commercial application, and a great worked example
- Query Understanding in Practice — the pipeline that turns a raw query into intent
- Search Quality Assurance — how you know whether search is any good
- A-B Testing for Search — measuring change in production
- Relevance Program Setup — the methodology that ties it all together
- Search UX — search is a product problem, not only a technical one
Then branch into whatever vertical interests you: Enterprise Search, Multilingual Search, Conversational and Agentic Search, Personalization in Search.
2. Move to the Concepts
Once a topic raises a term you don’t know, follow it into Concepts. Don’t try to read all 130+ — learn this foundation first, in roughly this order:
Retrieval & scoring
- BM25 — the lexical baseline you must understand cold
- Full-Text Search · Precision and Recall — the vocabulary of every interview
- Semantic Search · Dense Vector Retrieval · Embeddings
- Hybrid Search · Reciprocal Rank Fusion — how lexical and vector retrieval combine
Ranking
- Learning to Rank · LTR Feature Engineering · Reranking
- Position Bias · Click Signals — why naive use of clicks misleads
Evaluation (the part most candidates are weakest on)
Query understanding
- Query Understanding · Query Types · Search Intent
- Synonyms · Spelling Correction · Query Expansion · Zero Results
If you only memorize one thing for interviews: be able to explain BM25, the difference in Precision and Recall, and how you would measure whether a ranking change helped (NDCG + Judgment Lists).
3. Follow People and Companies
Search has a small, unusually open community. Following the right people gives you a free, continuous curriculum.
Educators & writers to follow first
- Daniel Tunkelang — canonical writing on relevance, query understanding, and search teams
- Doug Turnbull — Relevant Search author, relevance methodology, prolific blogger
- Trey Grainger — AI-Powered Search author, neural/semantic search
- Charlie Hull — vendor-neutral strategy, Haystack organizer, community builder
- Jo Kristian Bergum — deep, practical retrieval and ranking writing (co-founder of Hornet; formerly Vespa)
- James Rubinstein — search evaluation and metrics
- Giovanni Fernandez-Kincade · Atita Arora · Nikhil Dandekar · Eugene Yan — applied search & ML
See the full roster in People.
Companies whose engineering blogs are worth tracking
- Consultancies that publish openly: OpenSource Connections, Sease, The Search Juggler
- Platform vendors: Elastic, Vespa, Algolia, Weaviate, Qdrant, Cohere
- End-user teams with strong case studies: Etsy, Airbnb, Uber, Spotify, Zalando, Reddit
The Case Studies are how these companies actually solved real problems — read them like field reports.
4. Read the Books
The full annotated list is in Books. To start:
- Free, foundational: Introduction to Information Retrieval (Manning, Raghavan, Schütze) — the standard text, free online
- Most practical first read: Relevant Search (Doug Turnbull, Berryman) — mental models for relevance tuning
- Modern / neural: AI-Powered Search (Trey Grainger et al.) — semantic and neural search in practice
Read Relevant Search for intuition, Introduction to IR for rigor, AI-Powered Search for where the field is now.
5. Read the Case Studies
Books and courses teach the principles; case studies show how the industry leaders actually applied them — with the constraints, dead-ends, and tradeoffs textbooks omit. The Case Studies folder collects how real teams built and fixed search at scale. Read each like a field report and ask: what was the problem, what did they try, how did they measure success, and what would I have done differently?
| Case study | Company | What to learn from it |
|---|---|---|
| Uber Eats - Scaling Search for Food Delivery | Uber | Geo-constrained retrieval and ranking at scale |
| Airbnb - ML-Powered Experiences Ranking | Airbnb | ML ranking and the data flywheel |
| Etsy - Search Quality and Query Understanding | Etsy | Query Understanding in a long-tail marketplace |
| Skyscanner - Learning to Rank for Flights | Skyscanner | Learning to Rank in production |
| Slack - Enterprise Message Search with LTR | Slack | Enterprise/personal search with LTR |
| Netflix - Content Search Architecture | Netflix | Search Architecture for a content catalog |
| Canva - Search Pipeline Modernization | Canva | Modernizing a legacy search pipeline |
| Reddit - Vector Database Selection | How to actually choose a vector database | |
| Kleinanzeigen - Vespa Migration for Homepage Feed | Kleinanzeigen | Migrating to Vespa for feed ranking |
| Zalando - Self-DoS via Facet Aggregation | Zalando | A cautionary failure: facets that took search down |
The Zalando story is a reminder that the most instructive case studies are often the failures. Browse the full set via MOC - Case Studies.
6. Take a Course (Even a Free One)
A structured course gives you scaffolding the blog-and-book path lacks. Full annotated list: Courses. Strong options, many free:
- Cheat at Search Essentials (Doug Turnbull) — free, beginner-friendly intro to retrieval; the ideal first structured exposure. See Courses
- AI-Powered Search: Modern Retrieval for Humans & Agents (Trey Grainger & Doug Turnbull) — paid 4-week cohort, the applied next step; cohort version of the AI-Powered Search book
- Stanford CS276 / CMU 11-642 Search Engines — university IR course materials are freely available online
- Coursera — Text Retrieval and Search Engines (UIUC, ChengXiang Zhai) — auditable for free
- OpenSource Connections — Think Like a Relevance Engineer (TLRE) — the practitioner-standard relevance training (OpenSource Connections)
- Vector-DB academies — free, hands-on intros from Weaviate, Qdrant, and Pinecone
- DeepLearning.AI short courses — short, free modules on embeddings, RAG, and retrieval
Pair any course with a small project (see the tools step) — the course teaches concepts, the project builds the evaluation instinct interviewers test for.
7. Attend the Conferences
Conferences are where you meet the community and see the current state of the art. Full guide: Events and Conferences.
Best for someone starting out (practitioner-focused):
- Haystack — the open-source search relevance conference; the most welcoming entry point
- Berlin Buzzwords — strong open-source search engineering track
- MICES — dedicated e-commerce search
- Activate · Elastic{ON} — vendor ecosystems
Academic (for depth / research roles): SIGIR, ECIR, TREC, WSDM, CIKM.
You don’t have to travel — many talks are posted free on YouTube, and there are active Slack/meetup communities (London Search & AI Meetup, Relevance Slack). Watching last year’s Haystack talks is a free crash course.
8. Explore the Tools
Theory sticks only when you build. Stand up a real engine, index a dataset, write some queries, and — crucially — measure your results. The Tools folder covers the landscape; compare options in Search Platforms.
Get hands-on with:
- A search engine: Elasticsearch or OpenSearch (compare in Elasticsearch vs OpenSearch); or Vespa for ranking depth
- A vector database: Qdrant Vector DB, Weaviate Vector DB, or Milvus Vector DB
- Evaluation tooling: Quepid — build Judgment Lists and measure relevance like the pros do
- Query tuning: Querqy for rules-based query rewriting
- LTR: RankLib with XGBoost / LightGBM / CatBoost models
- Even your existing database: Search using PostgreSQL with pgvector is a low-friction starting point
A good first project: index a public dataset (products, Wikipedia, papers), build a small judgment list, measure NDCG for a baseline BM25 query, then try to beat it with synonyms, then with Hybrid Search. That single project demonstrates retrieval, ranking, and evaluation — exactly the take-home relevance task many teams use to hire.
9. Get Active in the Community
Following people (step 3) is passive; participating is what actually builds a network, surfaces job leads, and accelerates learning. Search is small and welcoming enough that newcomers who ask good questions get noticed.
Relevance Slack — the central hub
The Relevance Slack community, created and run by OpenSource Connections, is the single best place to be active. Practitioners share tips across Solr, Elasticsearch, Learning to Rank, vector search, and product management — and post jobs.
🔗 Join: https://opensourceconnections.com/slack
Notable channels:
#jobs— search-specific job opportunities (a real hiring channel, not a board)#jobs-eu— Europe-focused search roles#blogs-papers-books— curated resource sharing#es-learn-to-rank— Elasticsearch Learning to Rank discussion#quepid— questions about Quepid
Where to actually find a search job: the #jobs and #jobs-eu channels in Relevance Slack are the field’s de-facto job board — niche search/relevance roles get posted here that never reach LinkedIn. Lurk them, but also be active in the topic channels first: many openings are filled by referral from people who’ve seen you ask good questions.
Norms: vendor pitches are discouraged; the tone is polite and supportive, oriented toward weighing the pros and cons of technical options. Background: “Building the Search Community with Relevance Slack”.
Ways to actually participate
- Answer questions in Slack on things you’ve just learned — teaching cements knowledge and builds reputation
- Write in public — blog your first-project results (from step 8); a single honest “I measured BM25 vs Hybrid Search and here’s what happened” post is more credible than a résumé line
- Show up at meetups — London Search & AI Meetup and local equivalents; talks from Events and Conferences often have associated communities
- Contribute to open source — issues and docs for Quepid, Querqy, Elasticsearch, or a vector DB are low-barrier entry points
- Go to conferences and talk to people — Charlie Hull, Doug Turnbull and other organizers are notably approachable
Building in public + being genuinely helpful in Slack is, for many people, how the first search job actually arrives — through the
#jobschannel or a referral, not a cold application.
10. Monitor Where the Frontier Is
Steps 1–9 get you to today’s baseline — but the baseline keeps shifting, and what’s table-stakes one year is assumed knowledge the next. Staying current has to be a habit, not a one-off. This vault keeps the map: Frontier of Search, a year-by-year index of the leading edge, anchored in time so it reads as a dated snapshot rather than a vague buzzword. Start there and follow it into the period pages — Frontier of Search 2025, Frontier of Search 2026, and whatever year is current.
Skim each year’s page and diff it against the last. The skill you’re training is the one interviewers and clients pay for: telling a durable shift from hype that won’t outlast a news cycle.
Roles in Search
Know which job you are aiming for — the skills overlap but the emphasis differs. Full breakdown, skills, and “what separates good from great” in Hiring for Search.
| Role | Owns | Core skills to build | Best starting background |
|---|---|---|---|
| Relevance / Search Engineer | The bridge from IR to production | BM25, Learning to Rank, Query Understanding, engine internals, Search Evaluation | Backend / software engineering |
| ML Engineer (Ranking/Retrieval) | The model layer | LTR Feature Engineering, embeddings, Reranking, vector search, Click Models | ML / data engineering |
| Search Data Scientist | The measurement layer | A-B Testing for Search, Position Bias, query-log analysis | Statistics / data science |
| Search Product Manager | Strategy & roadmap | Technical literacy, measurement orientation, stakeholder translation | Technical PM |
| Relevance Annotator / Judgment Analyst | Judgment Lists & eval ground truth | Calibration, annotation guidelines, domain knowledge | Domain expertise; a common entry point |
| Search Consultant | Diagnosis & advisory across clients | All of the above + communication | Experienced practitioners — see Search Consultancy |
Easiest entry points for someone breaking in: starting as a backend engineer on a team that owns search, a judgment/annotation role that grows into relevance engineering, or an ML role adjacent to ranking. The annotator and data-science paths are often underrated doors into the field.
Two Career Tracks: Employment vs. Consultancy
The roles above describe an employed track — you join a team that owns search. There is a second track: independent consulting / advisory, where you sell search expertise to many clients. It’s covered in depth in Search Consultancy.
| Employment | Consultancy | |
|---|---|---|
| Where the work comes from | #jobs / #jobs-eu in Relevance Slack, referrals, company boards (up in step 9) | Inbound from your public profile, referrals, network overflow, RFPs |
| What you sell | Your time to one employer | Diagnosis + roadmap + hands-on work to many clients |
| Prerequisite | IR fundamentals + evaluation mindset | The above plus a track record and a visible reputation |
| Best first move | Land an employed search role; build depth | Don’t start here cold — most consultants build credibility employed first |
Consulting is rarely a starting point — it’s where experienced practitioners go after building a reputation. The path to it is the same community work in step 9: writing in public, speaking at conferences, and contributing to open source are what generate inbound consulting leads. Search Consultancy details positioning, engagement models (day-rate → fixed-scope → retained), and using your network as pipeline.
Related
- Hiring for Search — what teams screen for (read this to reverse-engineer how to prepare)
- Managing a Search Team — where these roles sit once you’re in
- Search Consultancy — the independent / advisory path
- Economics of Search — why search teams exist and how they’re justified
- Books · Courses · Events and Conferences · Search Platforms — the companion reference lists