Information Extraction Engineer

About the role

ABOUT DEFCON AI

RESILIENCE IN THE FACE OF DISRUPTION. DEFCON AI is an insights company that leverages artificial intelligence, mathematical optimization, data analytics, and software engineering for resilient optimization of complex systems. In today’s dynamically changing world, DEFCON AI’s technology aligns outcomes with operational goals, better decision making, and empowers customers to anticipate assess, and mitigate the impacts of disruptions.

About the Role

You’ll pull structured facts out of long, free-text documents, and keep every one of them tied to the exact passage it came from. On this platform, nothing gets asserted without a citation a human reviewer can check, and that traceability is yours to build and protect.

You’ll join the analytics and AI engineering team behind a system that ingests records from dozens of disparate sources, resolves them to the right entity, and shows a human reviewer the exact source text behind every answer. Operating within a secure government cloud environment, the platform depends on extraction a reviewer can trust precisely because it can be checked against the original document.

You’ll join an established extraction practice rather than building one from scratch. The approach is set and reviewed by a senior practitioner on the team. Your contribution is sustained, consistent output at volume against long, messy, narrative documents that were never written with extraction in mind, and the judgment to flag what doesn’t fit rather than force it.

This is a fully remote role, with a later start date than our most urgent openings. We’re sourcing now so the seat is filled when the work begins.

Key Responsibilities

- Extract structured facts from long, free-text, narrative, and semi-structured documents

- Maintain a traceable pointer from every extracted fact back to the exact source passage it came from

- Validate that every cited passage actually supports the extracted fact, and treat a missing or unsupported citation as a defect

- Distinguish a stated total from a derived sum in financial narratives, retaining the contributing items and the calculation for verification

- Consume the recorded rule context supplied to model-assisted extraction, keeping cited facts and uncertainty separate from any validity decision

- Apply and extend the team’s established extraction approach consistently across new document types

- Handle ambiguous, contradictory, or incomplete source text without forcing a single reading; flag it rather than guessing

- Work with entity resolution and retrieval teammates on what a usable extracted fact needs to carry downstream

Required Qualifications

- 3+ years extracting structured facts or data from unstructured, narrative text in a production setting

- Experience keeping extracted output traceable to its source

- Strong Python, with hands-on NLP or information-extraction tooling, whether rule-based, statistical, or LLM-based