ML Engineer, Retrieval & Grounded Generation
$165k - $200k • Remote • Remote, USA • Full-time
Posted 1mo ago
Job Location
Remote, USA
Tech Stack
Remote Work Policy
Fully remote
Employment Type
Full-time
Visa Sponsorship
The posting says they won't sponsor visas
Categories
AI Engineer
About the job
DEFCON AI is seeking an ML Engineer to join their analytics and AI engineering team. You will work on an AI-assisted platform that integrates records from numerous external feeds, resolves them to the correct person, highlights key information for human reviewers, and provides clear, defensible explanations for every recommendation. This role operates within a secure government cloud environment, emphasizing the importance of accuracy and detail in a system that addresses critical challenges.
As an ML Engineer specializing in Retrieval & Grounded Generation, you will be responsible for developing embeddings, vector storage, and retrieval systems at scale. You will integrate language models to ensure all generated text is linked to cited source records and rigorously test for citation failures. Your duties will also include designing prompts and output schemas, managing model packaging, versioning, serving, and rollback procedures, and implementing telemetry hooks for effective measurement and analysis of the production system.
Responsibilities
- Implement embeddings, vector storage, and retrieval for a large, provenance-tracked evidence base.
- Integrate language models to ensure generated text is bound to cited source records and test for citation failures.
- Design prompts and output schemas.
- Manage model packaging, versioning, serving, and rollback.
- Instrument telemetry for retrieval and generation quality, recommendation attribution, overrides, abstentions, grounding failures, latency, and throughput.
- Provide bounded model assistance for narrative extraction where deterministic processing is insufficient, ensuring outputs are tied to source passages.
- Supply recorded rule context to model-assisted steps, ensuring outputs carry exact rule versions and ordered context.
- Maintain a modular in-boundary serving path, self-hosted or managed, to avoid dependency on a single provider's availability.
Requirements
- 5+ years of experience, including building a production or near-production retrieval-augmented (RAG) system.
- Detailed understanding of retrieval design, vector store selection, grounding testing, citation failure analysis, and rollback procedures.
- Strong Python skills with hands-on experience in embeddings and vector retrieval at scale.
- Demonstrated clarity on shipped work (prototype, proposal, or deployed code).
- US Citizenship required.
- Active US Secret clearance required to start.
Benefits
- Competitive salary, bonus, and equity package
- 100% employer paid, comprehensive health insurance (medical, dental, vision) for you and your family
- Unlimited PTO with manager approval
- Flexible work environment
- 14 weeks of fully-paid parental leave
