AI Engineer

Marathon TS is hiring an AI Engineer for its Risk Decision Group to advance LLM and document-intelligence capabilities on a greenfield data and AI platform built for a high-trust federal environment. This is a hands-on, near-term product demonstration effort with documentation and transfer to an internal team over a 1-year contract (with an option to extend). Active T5/SSBI clearance and U.S. citizenship are required.

What You’ll Do

• Build RAG and document-processing pipelines on the Databricks lakehouse, including ingestion, OCR for mixed-quality sources, chunking, embedding, and retrieval.
• Create LLM workflows for summarization, structured extraction, and evidence-grounded generation with source attribution.
• Develop synthetic document corpora with fidelity and quality variation needed to produce meaningful results.
• Stand up an evaluation harness covering retrieval quality, groundedness and hallucination checks, structured-output validity, and human-in-the-loop review, and report results in numbers.
• Package deliverables as jobs and Asset Bundles, track work in MLflow, and document what the internal team needs to own the system.

Core Requirements

• U.S. citizenship and active T5/SSBI federally adjudicated clearance.
• 8+ years building applied ML/AI or data systems, with demonstrated delivery of LLM and RAG systems you personally built (not notebook demos).
• Hands-on Databricks.
• Document processing at scale, including OCR, layout-aware parsing, chunking tradeoffs, and handling poor-quality sources.
• Local or self-hosted LLM serving (vLLM, TGI, Ollama, llama.cpp, or equivalent), including running open-weight models in an isolated or air-gapped environment without reliance on external API endpoints.
• Structured extraction and grounded generation with source attribution.
• Experience with LLM evaluation methodology, including how you measured correctness and what the evaluation missed.
• Privacy-preserving synthetic data generation from CUI, PII, or comparably restricted source data, including understanding re-identification risk.
• Strong Python.
• Government or defense contracting experience.

Technologies

• Databricks, Databricks lakehouse, RAG, OCR, embeddings, LLM
• vLLM, TGI, Ollama, llama.cpp
• MLflow, Asset Bundles

Preferred Qualifications

• RAG built inside a government or FedRAMP-authorized environment (examples listed: Azure OpenAI in GCC High, AWS GovCloud, Bedrock within an authorized boundary).
• Experience handling FedRAMP Moderate, NIST 800-171, CMMC L2, or CUI.
• Databricks capabilities and tooling including Databricks Vector Search, Mosaic AI Agent Framework and Agent Evaluation, Asset Bundles, and MLflow.
• Experience with Unity Catalog governance.
• Experience with H2O (h2oGPTe, Driverless AI).

Security, Location, and Pay

• Security clearance: Top Secret (required)
• Work location: Remote
• Pay: USD 82 to 92 per hour
• Job type: Contract

Application Screening Questions

• Do you have at least 8 years of building applied ML/AI or data systems, with demonstrated delivery of LLM and RAG systems you personally built (not notebook demos)?
• Do you have document processing at a scale involving OCR, layout-aware parsing, chunking tradeoffs, and poor-quality source handling?