Exa (formerly Metaphor Systems) is a San Francisco company building a neural search API aimed at AI applications and agents rather than human readers. It was founded in 2021 by Will Bryk and Jeff Wang and is backed by Y Combinator. Instead of returning ten blue links, Exa returns full page content ranked for what an AI model needs, through a web search API, a crawling API, and a SERP API used by companies such as Cursor, Cognition, and HubSpot along with hundreds of thousands of developers. The funding path has been steep: a $22M Series A led by Lightspeed in 2024, an $85M Series B at a $700M valuation led by Benchmark in September 2025, and a $250M Series C at a $2.2 billion valuation led by Andreessen Horowitz in May 2026.
Process
Public detail on Exa’s loop is thin, so treat this as a picture drawn from a few Glassdoor accounts and the company’s early stage, not a fixed script. Expect a short cycle, often one to two weeks from first contact to decision, which is much faster than a big-company loop. Reports describe it opening with a brief, fast-paced technical screen of roughly 15 minutes that mixes quick computer-science facts, a probability question, and a high-level design prompt. That is usually followed by a take-home project and then a full trial workday where you build something and present it to the team at the end. Sharpen your coding patterns so the screen goes smoothly, be ready for system design centered on indexing and retrieval, and prepare a couple of behavioral stories about shipping fast when the spec is loose.
What they actually ask
- Design a search and ranking system: how would you combine keyword matching with embedding-based retrieval, rank results for a query, and keep latency low?
- Vector search internals: how do you build and scale an approximate nearest neighbor index over billions of vectors, and what are the recall-versus-latency tradeoffs?
- Crawling and indexing: design a web crawler that stays fresh, deduplicates pages, respects politeness rules, and extracts clean text from messy HTML.
- Latency and infrastructure: how would you serve sub-second search across a large index using sharding, caching, and the right split between GPU and CPU for embedding inference?
- Retrieval for AI: what makes a result “good” for an LLM or agent rather than a human, and how would you evaluate and improve that quality?
- A practical build task from the take-home or trial day: stand up a small service or feature end to end and explain your tradeoffs.
Levels and comp (2026)
Exa does not publish salary bands, and as a private, venture-backed company its numbers shift with each round, so treat any figure as a rough guide and check levels.fyi and recent Blind threads for current bands before you negotiate. Expect the usual startup mix: a competitive base that is typically below top FAANG cash, a meaningful equity grant, and standard health benefits. For senior software engineers in San Francisco, cash bases at well-funded AI startups commonly sit somewhere from the low-to-mid $100Ks up toward $200K or a bit higher, with equity as the real lever. That equity is where both the upside and the risk live. After a $250M Series C at a $2.2 billion valuation, the grant is priced off a high number, so options can pay off well or come to little depending on the outcome. Ask about the share count, the current preferred price, the vesting schedule and cliff, and any early-exercise or extended post-termination exercise window before you sign.
Prep priorities
Start by using the product. Read Exa’s API docs, run some searches, and get a feel for how results differ from a normal search engine, because that gives you concrete material for the trial day and shows real interest. Drill retrieval fundamentals next: embeddings, vector indexes and approximate nearest neighbor search, BM25 versus semantic ranking, and evaluation metrics like recall, precision, and nDCG. Practice system design that is specific to search infrastructure, including crawling pipelines, index sharding, caching, and latency budgets, rather than generic “design Twitter” prompts. Keep your coding fast and clean, but do not over-invest in hard algorithm puzzles; this is a build-quickly shop, so practice turning a loose problem into working code under time pressure. Finally, rehearse for the trial day itself: scope a problem tightly, ship a working slice, and be ready to explain what you cut and why.
Frequently Asked Questions
Does Exa run hard LeetCode interviews? Based on public reports, the loop leans toward a take-home and a trial day where you build real things, with less weight on hard algorithm puzzles than a big-tech loop. Keep coding sharp, but expect practical work over whiteboard trivia.
How long does the process take? It is short. Candidate reports point to roughly one to two weeks from first contact to a decision, faster for full-stack roles and a little slower for specialized ones like web crawler engineer.
What is the trial day? It is a full working day, often paid, where you take on a real problem and present your solution to the team. Treat it like a short onboarding: scope tightly, get something working, and communicate your tradeoffs clearly.
Do I need machine learning or search experience? It helps but is not always required, depending on the role. Full-stack and infrastructure roles reward strong engineering and fast learning; understanding retrieval basics and having actually used Exa’s API will set you apart either way.
Similar company guides
Prepping for Exa? Put it to work:
