About the Role :

We are looking for a Senior AI Engineer to build a production-grade Document Intelligence Platform powered by AI Agents and Amazon Bedrock. The platform will ingest PDFs of varying quality, extract structured information into predefined schemas, generate field-level confidence scores, and support Human-in-the-Loop (HITL) review workflows.

The ideal candidate should have strong expertise in AI Agents, Document AI, OCR/VLM technologies, RAG architectures, and production-scale LLM applications on AWS.

Key Responsibilities :

- Design and develop AI agents that extract structured data from unstructured and semi-structured documents.

- Build document ingestion pipelines capable of handling low-quality scans, image-based PDFs, and complex layouts.

- Develop confidence-scoring and routing mechanisms to automatically flag low-confidence fields for human review.

- Build and optimize RAG pipelines and domain-specific knowledge bases for grounded and accurate extraction.

- Implement Human-in-the-Loop workflows, reviewer feedback loops, and prompt-injection mitigation controls.

- Develop reprocessing and comparison mechanisms to identify material changes between document versions.

- Optimize model selection, inference costs, latency, and overall platform performance on Amazon Bedrock.

- Mentor engineers and establish reusable standards for extraction and orchestration frameworks.

Required Skills & Experience:

- 6+ years of software engineering experience with strong exposure to AI/ML and Generative AI applications.

- Hands-on experience with AWS and Amazon Bedrock, including prompt engineering, model evaluation, and production optimization.

- Strong experience with AI Agent frameworks such as LangGraph, LangChain, or similar orchestration platforms.

- Expertise in OCR technologies (Amazon Textract or equivalent), Vision Language Models (VLMs), and Document AI solutions.

- Experience building production RAG systems, vector retrieval, chunking, and grounding techniques.

- Proven experience extracting structured data from large PDF documents with high accuracy.

- Strong understanding of confidence estimation, Human-in-the-Loop systems, and evaluation frameworks.

- Experience with Python and production-grade LLM pipelines.

Good to Have :

- Experience with confidence calibration and selective prediction techniques.

- Exposure to Firecracker/microVM sandboxing for untrusted documents.

- Experience with evaluation harnesses, golden datasets, and drift monitoring.

- Knowledge of Commercial Real Estate (CRE) data structures and governance practices.