Senior Machine Learning Engineer

Company Overview Pantheon Data (a Kenific Holding company) is a private, small business based in the Washington, DC, area. Pantheon Data was founded in 2011, initially providing acquisition and supply chain management services to the US Coast Guard. Our service offerings have grown in the past ten years, including infrastructure resiliency, contact center operations, information technology, software engineering, program management, strategic communications, engineering, and cybersecurity. We have also grown our customer base to include commercial clients. The company has used this experience to expand our service offerings to other agencies within the Department of Homeland Security (DHS), the Department of Defense (DoD), and other Federal Civilian Agencies. Position Overview Pantheon Data is seeking a Senior Machine Learning Engineer to design, build, and operate production AI systems for federal clients - including hybrid retrieval-augmented generation (RAG) applications, Intelligent Document Processing (IDP) pipelines, and LLM-backed decision-support tools running in AWS GovCloud. This is a hands-on, production-focused engineering role. You will work with GovCloud-hosted foundation models on Amazon Bedrock, build hybrid retrieval over relational and vector data stores, design rigorous evaluation harnesses, and ship secure, traceable, evidence-grounded systems in FedRAMP High / DoD IL4–IL5 environments. This is not a notebook-only, prompt-only, or research-only role: successful candidates can walk through real systems they have built - the data flow, retrieval and inference architecture, deployment approach, evaluation strategy, failure modes, and what they personally implemented. What This Role Will Work On Design and build hybrid RAG applications combining semantic (vector) search, lexical/full-text search, and structured SQL retrieval, with rank fusion, evidence grounding, and source-cited answer generation. Integrate and operate GovCloud-hosted LLMs and embedding models via Amazon Bedrock (e.g., Anthropic Claude, Amazon Titan embeddings), including prompt/version management, guardrails, streaming inference, and cost/latency tuning. Build Intelligent Document Processing capabilities: OCR post-processing, layout-aware parsing,tableand form extraction, NLP/LLM structured extraction, chunking and embedding strategies, and human-in-the-loop review workflows. Develop serverless, event-driven ingestion pipelines (AWS Step Functions, Lambda, S3, Aurora PostgreSQL withpgvector) that turn messy source documents into reliable, traceable,queryableinformation. Design and run evaluation programs for retrieval quality, extraction accuracy, answergroundedness, hallucination risk, and end-to-end system performance - including golden datasets, automated eval harnesses in CI, and regression testing across model and prompt versions. Engineer for security and compliance from the start: least-privilege IAM, private VPC endpoints and restricted-egress network boundaries, guardrails and prompt-injection defenses, output validation, audit logging, and CUI-aware data handling in support of ATO requirements. Make practical engineering decisions about when to use deterministic logic, classical NLP, OCR, embeddings, LLMs, fine-tuned or distilled models, or humanreview -and defend those tradeoffs to technical and non-technical stakeholders. Responsibilities Design, implement, andmaintainML/AI software components for RAG, IDP, and generative AI systems serving federal customers. Build andoperatedata pipelines for unstructured and semi-structured data: ingestion, extraction, cleaning, enrichment, validation, quality scoring, quarantine/review, and storage. Develop and evaluate NLP, OCR, computer vision, embedding, retrieval, and LLM-based approaches for document understanding and question answering. Contribute production-quality Python with clear structure, tests, logging, error handling, and documentation;participatein code review and team-based delivery. Deploy and support AI services in cloud and containerized environments, including Bedrock model integration, batch processing, workflow orchestration, and observability (CloudWatch, tracing, structured logs). Define evaluationmethodologyand metrics; build eval datasets and harnesses; analyze errors; and drive iterative improvement of retrieval and generation quality. Implement responsible-AI and security controls: Bedrock Guardrails or equivalent, PII/sensitive-data redaction, content filtering, prompt-injection mitigation, and output traceability to source evidence. Troubleshoot system behavior across model output, data quality, retrieval, schema design, infrastructure, latency, cost, and user workflow. Mentor other engineers,helpestablishstandards for evaluation, reproducibility, and responsible AI use, and raise the technical quality of the team. Communicate clearly with technical and non-technical stakeholders, inc