AI Relation Extraction Pipeline Developer

Altrix Systems

Remote · $100k – $130k / year · Full-time · Posted 4 months ago

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

About the role

A knowledge graph is only as good as the relations that connect its entities. We have 14 million nodes; the edges between them — who acquired whom, which drug targets which protein, which regulation applies to which jurisdiction — are what make the graph useful. Most of those edges are buried in unstructured text, and we're extracting maybe 40% of the relations that actually exist in our document corpus. You'll improve the relation extraction pipeline that populates our graph edges. The challenge is coverage without noise: precision matters because a wrong edge corrupts downstream reasoning, but recall matters because missing edges leave analysts with an incomplete picture. You'll build joint entity-relation extraction models, design few-shot learning approaches for new relation types, and implement the confidence scoring that determines which extractions are trustworthy enough to add to the graph automatically versus which need human review. The NLP team is five engineers. You'll own the extraction pipeline end-to-end, collaborating with the graph team who maintains the schema and the domain experts who validate extractions in specialized fields.

Fully remote.

Responsibilities

Build joint entity-relation extraction models improving coverage from 40% to 75%+
Design few-shot learning approaches for new relation type onboarding
Implement confidence scoring to route low-confidence extractions for human review
Create evaluation benchmarks with domain-expert-labeled relation samples
Build the review interface that domain experts use to validate and correct extractions
Monitor extraction quality per relation type and document domain

Requirements

3+ years in NLP with information extraction or relation extraction experience
Strong PyTorch and HuggingFace skills
Experience with few-shot learning and low-resource NLP techniques
Understanding of knowledge graph schemas and property graphs
Ability to work with domain experts across legal, biomedical, and financial fields