# Software Engineer, Search Systems - Code Data at Mercor

- Company: Mercor
- What the company does: Mercor is organizing human intelligence to power the AI economy. We are powering frontier research, AI benchmarks, and AI agent training at scale for the top AI labs and enterprises. Backed by General Catalyst and Menlo.
- Company website: https://mercor.com/
- Type: Startups
- Level: Mid level
- Location: San Francisco
- Work setup: On-site
- Pay: $250K to $500K base salary per year (USD)
- Posted: 2026-08-14
- Apply by: 2026-10-08
- Apply: https://jobs.ashbyhq.com/mercor/aac13ab1-a746-437a-8c56-daf6e8b6c542
- Page: https://www.1752.vc/careers/jobs/mercor-software-engineer-search-systems-code-data/

## About the role

Some of the most valuable work on Mercor's platform is code —the tasks, problems, and solutions that train and evaluate the world's frontier coding models. As a Software Engineer for Code Search & Retrieval , you'll own the architecture and algorithms behind how we search across code tasks: finding similar tasks, routing them to the right models, and turning natural-language questions into precise retrieval.

## What they're looking for

- 8+ years of professional software engineering experience, including 3+ years operating at a Senior level or above, with a Staff-level track record of org-wide technical impact
- Deep, hands-on expertise building search and retrieval systems : dense-embedding retrieval, lexical scoring (BM25/TF-IDF), hybrid ranking, and re-ranking
- Good to have but not required: Experience with code search or code understanding —retrieval over code, code embeddings, or working with code-specific models—and an appreciation for why matching similar code tasks is harder than matching text
- Strong understanding of the search algorithms and index internals —vector/ANN indices (e.g. HNSW, IVF, product quantization), inverted indices, and engines such as Elasticsearch/OpenSearch, Lucene, FAISS, or vector databases
- A track record of making the right cost-vs-speed tradeoffs : latency budgets, throughput, memory footprint, and infrastructure spend on high-QPS systems
- Familiarity translating natural-language questions into structured search queries (query understanding, semantic parsing, or LLM-assisted query generation)

Tags: Engineering