ML Engineer — Legal Document Intelligence

Ethan Caldwell

Contract-to-hire · Mid-level · New_York

This role is no longer accepting applications

Similar open roles you can apply to now:

About the role

We build e-discovery software used by litigation support teams at large law firms and corporate legal departments. Our platform ingests document collections ranging from tens of thousands to tens of millions of files per matter, applies classification and relevance scoring, and surfaces material that attorneys need to review. We have been doing this since 2012. The rules-based classification system we have used since 2014 is adequate. It is not good. Its accuracy degrades on novel document types, it requires manual tuning per matter that takes our customer success team significant time, and it does not learn from review decisions made by attorneys on the platform. We know what we need to build. We need a machine learning engineer who can build a document classification and relevance scoring system that replaces the rules-based layer, integrates with our existing document processing pipeline, and improves in accuracy as attorney review decisions accumulate. The work is technically interesting. The domain is niche. The quality bar is high because inaccurate classification in litigation support has real consequences. This starts as a six-month contract — we want to make sure the fit is right and the technical approach is sound before converting to a permanent hire, which is our strong intention if both conditions are met.

Responsibilities

Design and implement a document classification and relevance scoring system to replace the current rules-based layer
Build active learning or feedback integration mechanism so that attorney review decisions improve model accuracy over time
Develop evaluation frameworks that measure performance in terms meaningful to legal review workflows — privilege recall, responsiveness precision, and review efficiency
Work with our document processing team to integrate the ML components cleanly into the existing pipeline
Document the system architecture, model behaviour, and evaluation methodology clearly enough to support customer due diligence conversations

Requirements

4+ years of ML engineering with experience in document classification, information retrieval, or text processing systems
Hugging Face for fine-tuning document classification models on domain-specific legal text
Python for model training pipelines, evaluation tooling, and integration with existing document processing infrastructure
SQL for working with the document metadata and attorney review decision data that feed the training and evaluation pipeline
RAG or embedding-based retrieval experience — the relevance scoring component will use dense retrieval over document embeddings
LlamaIndex or equivalent for building the retrieval and ranking layer
Experience with precision-recall trade-offs in high-stakes classification contexts — legal e-discovery has specific recall requirements that differ from typical classification tasks

Benefits

$118,000 – $142,000 annualised contract rate
6-month initial contract with strong intention to convert to permanent hire
Fully remote — our engineering team is distributed across the US and Canada
Domain training provided — you will have access to attorneys and litigation support specialists who will explain what good classification actually means in practice
Equity conversation at permanent conversion if the engagement goes well