Senior AI/ML engineer, document intelligence

Binariks is looking for an AI/ML Engineer, Document Intelligence with strong applied LLM and OCR experience to build the AI extraction core of our configurable insurance underwriting platform.

We are building a configurable AI product that automates document-heavy submission processing for underwriting teams—ingesting mixed-format documents, classifying them, extracting structured fields with source evidence, and surfacing completeness signals for human review. This is application-layer AI engineering: integrating and orchestrating OCR and LLM services into a production pipeline, not training or fine-tuning models.

We are looking for an engineer who can own the AI extraction pipeline end-to-end—from defining the data contracts it produces, through building the classification and extraction logic, to shipping evidence-linked results that a backend team turns into review screens. This role sits at the intersection of AI engineering, data architecture, and backend integration.

Your responsibilities

• Design canonical data contracts (submission, source document, extracted field) shared between the AI pipeline and the backend services that consume it.
• Build OCR and document preprocessing for mixed formats (scanned and native documents), using cloud document-intelligence services.
• Implement whole-document classification using LLMs, tuned to the client’s document types.
• Build structured field extraction that ties every extracted value back to its exact source location (evidence/provenance), not just the value itself.
• Design provider interfaces and adapters for OCR, LLM, and storage services that stay cloud-agnostic in principle while shipping against one cloud provider first.
• Implement non-binding completeness and attention signals (missing fields, type/range checks) without building calibrated confidence scoring or automated accept/decline logic.
• Set up lightweight, structured logging/tracing for LLM and OCR calls to make pipeline behavior debuggable, without building a full observability or evaluation platform.
• Integrate AI pipeline calls into an async, queue-based job architecture (e.g. FastAPI/Celery), working closely with backend engineers on retries, idempotency, and failure handling.’
• Run smoke-level validation against existing sample datasets to confirm the end-to-end extraction path works, ahead of underwriter UAT.
• Work with the backend/frontend team to make sure extracted data and evidence are structured in a way that supports human-in-the-loop review screens.

What We’re Looking For

• 5+ years of experience building applied AI/ML systems in production, ideally with a document-processing or data-extraction focus.
• Hands-on experience with LLM-based classification and structured extraction (prompt design, schema-constrained outputs, evaluating extraction quality on real documents).
• Experience with OCR/document-intelligence services (Azure Document Intelligence, AWS Textract, or equivalent) and preprocessing mixed-format documents (PDF, scans, Office formats).
• Ability to design clean data contracts/schemas that other teams (backend, frontend) can build against.
• Comfortable working inside a backend service architecture (FastAPI, async task queues) well enough to integrate AI calls into a production pipeline — this is not a pure backend role, but requires backend literacy.
• Pragmatic engineering judgment: able to ship lightweight, “good enough” solutions for logging, validation, and testing rather than defaulting to building full platforms (evaluation harnesses, observability dashboards, calibrated confidence models) before they’re needed.
• Experience working on a single-cloud-first, portability-minded architecture (interfaces/adapters designed for future multi-cloud, without over-building it upfront).
• Strong analytical skills — able to reason about document structure, edge cases, and failure modes in real-world business documents.
• Ability to work with ambiguous or evolving requirements typical of an early-stage product build.
• Upper-Intermediate or Advanced English.

Nice to have:

• Experience in insurance, underwriting, or another document-heavy regulated domain.
• Experience designing human-in-the-loop review workflows alongside a front