The Mission
At Unstructured, we are obsessed with transforming messy, unstructured data into a format that LLMs can actually use. Our Public Sector team has recently secured several high-impact contracts that demand more than just "off-the-shelf" solutions. We are looking for an AI Engineer who thrives at the intersection of R&D and production-grade software engineering.
You won’t just be building notebook demos; you will be architecting, prototyping, and shipping novel multimodal data processing, RAG, and agentic systems that solve critical problems for Government and Military clients. Your work will bridge the gap between one-off custom builds and a repeatable, scalable product roadmap.
What You'll Own & Drive
- You will be a high-agency individual contributor, owning the lifecycle of AI solutions from initial research to AWS deployment.
- 50% Building & Shipping: Design and implement production-grade RAG pipelines and agentic workflows using Python. You’ll build systems that handle real-world "messy" data (PDFs, scanned docs, images, full motion video) and ensure they are performant and scalable.
- 30% Research & Experimentation: Stay at the bleeding edge. You’ll evaluate new models (LLMs, embedding models, object detection), prototype approaches for SBIR/government deliverables, and run experiments to prove what actually works.
- 20% Strategy & Collaboration: Partner with the team to document architectures, contribute to technical reports for contract deliverables, and participate in pre-sales calls to architect solutions for complex client needs.
What We're Looking For
We are looking for a self-directed engineer who excels in high-stakes, ambiguous environments. You are likely a strong fit if your professional background reflects the following:
- Systems-First Engineering: You prioritize building reliable, scalable systems over experimental scripts. You have a track record of moving AI models out of notebooks and into production environments where latency, cost, and accuracy are treated as first-class citizens.
- Technical Resourcefulness: You are comfortable working in restricted or air-gapped environments. When commercial APIs aren’t an option, you have the expertise to deploy, fine-tune, and optimize open-source models to achieve the mission objective.
- Autonomous Problem Solving: You can take a high-level government requirement and translate it into a technical roadmap. You don't require constant oversight to identify the right tool for a job, whether it’s a specific vector database or a custom multimodal pipeline.
- A "Generalist" Mindset: While you specialize in AI, you understand the full stack. You are as comfortable discussing embedding strategies as you are configuring AWS GovCloud infrastructure or debugging a FastAPI endpoint.
Must-Haves
- Proven experience deploying Production RAG pipelines against real-world, messy datasets.
- Deep expertise in Agentic system design (tool-use, multi-agent orchestration).
- Strong Python engineering skills—writing clean, scalable, and maintainable code.
- Experience operating within AWS/GovCloud environments.
Nice-to-Haves
- Experience fine-tuning NLP or object detection models.
- Familiarity with LLM evaluation frameworks (hallucination detection, drift monitoring).
- Knowledge of government security standards and working in different classification environments and on-prem.
- Security Clearance: Existing Secret/TS clearance or eligibility is a significant plus.
Your Technical Toolkit
- Languages: Python (expert-level), SQL
- LLM & Agentic Frameworks: LangChain, LangGraph, CrewAI, or similar orchestration frameworks
- RAG Stack: Retrieval with vector databases (Pinecone, Weaviate, Chroma, pgvector), graph databases (Neo4J), Elasticsearch, BM25, and Sentence-Transformers; NLP enrichment with spaCy, GLiNER, and Transformers; optimization using embedding models, reranking pipelines, and DSPy
- Evaluation & Observability: RAGAS, DeepEval, Arize Phoenix, and synthetic annotations
- Cloud & Infrastructure: AWS, SageMaker, Bedrock, S3, Lambda, Docker, and FastAPI
- Data Processing: Complex pipelines for unstructured and multimodal data, including PDFs, scanned documents, images, and audio.