AI Engineer

Overview

Build the retrieval, matching and LLM systems behind JoVE's AI research products, owning features end to end from prototype and evaluation through production, monitoring and cost.

What you'll do

• Build hybrid retrieval (BM25, sparse and dense embeddings) with reranking over tens of millions of papers from PubMed, OpenAlex and PMC open-access full text.
• Improve how papers and protocol steps are matched to JoVE videos and timestamps, making matching aware of organism, technique and study type.
• Build RAG and agentic pipelines for cited synthesis, Methods Q&A and troubleshooting, where every claim traces to a source.
• Extract method steps, reagents and parameters from papers and link each step to the matching video chapter.
• Build eval sets with scientific editors, track relevance, citation accuracy, hallucination rate and match quality, and gate releases on them.
• Ship and run models and LLM services in production, watching latency and cost per query, and choose between hosted and open models.
• Help build the APIs and an MCP server that let AI assistants, ELNs and publisher widgets use JoVE's methods data.
• Work with product to measure the search, video click and watch funnel and use it to improve ranking.

What you'll need

• 4+ years in software or ML engineering.
• 2+ years shipping LLM, search or NLP features to production.
• Strong Python and solid software engineering habits: testing, code review, CI/CD.
• Hands-on experience building RAG systems: chunking, embeddings, hybrid retrieval, reranking and prompt design.
• Experience with a vector or search engine such as Elasticsearch/OpenSearch, Milvus, pgvector, Qdrant or Pinecone.
• Experience with LLM APIs (Anthropic, OpenAI or similar) and open models (Hugging Face, vLLM).
• A track record of building offline evals and using metrics, not demos, to decide what ships.
• Experience with agent frameworks, tool use, or building MCP servers.
• Comfort with cloud infrastructure (AWS), Docker and API design.
• Clear communication with non-engineers, including scientists and editors.

Nice to have

• Experience with scholarly data: PubMed, OpenAlex, Semantic Scholar, PMC, or entity extraction for genes, organisms and methods.
• Multimodal work: video transcripts, timestamp alignment, or image and video embeddings.
• Fine-tuning embedding or reranker models, or learning-to-rank.
• Experience with large-scale indexing (100M+ documents) and vector quantisation.

Details

• Fully remote.