Legal Engineer Intern — AI Quality & Evaluation 

Internship · Team: Machine Learning · [Paris — hybrid] · 6 months · Start: ASAP · Working languages: English & French 

About the role 

We are embedding AI across every DiliTrust module, and the hardest question is not "does it run?" but "is it good enough for a legal professional to rely on?" Answering that requires someone who understands both the model and the law. 

That is this role. You will join the ML team as the legal voice inside the build loop, translating legal expertise into the test sets, evaluation criteria and documentation that determine whether our AI features ship. You will work day to day with ML engineers and Product Managers, and your findings will directly shape what gets released and what goes back for rework. 

This is a legal engineering position, not a legal practice one. You will not draft contracts; you will define what a correct answer looks like on a contract, at scale, and hold the system to it. 

What you'll do 

Design and run AI evaluations (Lini) 

Build evaluation campaigns for Lini across all modules (CLM, Board Portal, Legal Entity Management, Matter Management). Define what "correct" means for each feature: scoring rubrics, acceptance thresholds, and the edge cases that matter to a lawyer but are invisible to a metric. Run the campaigns, analyse the results, and escalate quality issues to the ML team with a clear diagnosis rather than a bug report. 

Own the Golden Data 

Build, curate and maintain the reference test sets the team measures against. Source representative legal documents, establish the ground truth, and keep coverage honest as features evolve. This dataset becomes the team's working definition of quality, and it will be yours. 

Turn client feedback into product signal 

Collect and structure the feedback on AI features gathered by Customer Success. Build a failure taxonomy instead of a list of complaints, so that recurring weaknesses become prioritisable work items. Keep the feedback documentation current on Confluence. 

Map the competitive landscape 

Track legal tech products shipping notable AI capabilities and maintain a clear-eyed view of the market: what they claim, what they actually do, where we lead and where we don't. 

Document and enable 

Write and maintain the Confluence reference on our AI features — capabilities, limitations, appropriate use cases — for Product, Sales, Customer Success and Support. If a colleague can explain the limits of a feature to a client without asking the ML team, you have done this well. 

Who we're looking for 

A gap year student (Bac +4/+5) matching one of two profiles: 

• Legal background with a real pull toward technology. You have a law degree, and you find yourself more curious about how the tool works than about the case law. 

• Technical or engineering background with genuine interest in law. You can read a model output critically, and you want to work in a domain where precision has consequences. 

In both cases, the person we are looking for is: 

• Curious, autonomous and rigorous — able to own a workstream without being managed through it 

• Analytically minded: structures information, spots inconsistencies, synthesises clearly 

• An excellent communicator in French and English, written and spoken 

• Comfortable in a fast-paced environment where the roadmap moves 

Required skills 

• Solid legal knowledge, or a demonstrated ability to get up to speed on legal concepts fast 

• Fluent French and English — English is the working language inside the ML team 

• Proficiency with productivity tools: Confluence, Jira (or equivalent), Slack, Excel / Google Sheets 

• Nice to have: familiarity with LLMs, prompt engineering, or AI evaluation concepts. Hands-on experience with AI tools applied to legal work is a strong plus. 

What you'll get out of it 

Direct exposure to how AI features are built, measured and shipped in a production legal tech product — inside the ML team rather than adjacent to it. Legal engineering is becoming its own career track, and this is a year of it on a real product. 

Recruitment process 

Two steps, and we move quickly. 

1. Introductory call (20–30 min, phone or video) 

A conversation with [hiring manager / talent team] on your background, what draws you to this intersection of law and AI, and the practical basics (dates, duration, school requirements). Partly in English. 

2. Use case (take-home + debrief) 

We send you a short, concrete exercise: a set of AI outputs on legal documents to assess. You tell us what is wrong, how you would measure it systematically, and what you would report back to the ML team. Expect around [2 hours] of work. We then discuss it together for 45–60 minutes with the ML team and Product. We are not looking for the right answer — we are looking at how you reason, structure a problem, and defend a judgement call.