Senior Product Engineer - AI Native

Ready to be part of the Legal Tech revolution? Vision: As a leading software-as-a-service (SaaS) provider, DiliTrust is a global company dedicated to offering an integrated suite of legal and governance products. Our vision is to digitize legal departments worldwide. With an annual growth rate of over 40% since 2020, our ambition is to become the world's leading Legal Tech company, aiming for a valuation exceeding $1 billion by 2026. Our Impact: From generating General Meeting reports to leveraging AI-assisted contract lifecycle management, our teams in our 8 offices across France, the US, Mexico, MEA, Germany, Spain, Italy, and Canada are the driving force behind our global success. We proudly support 2,400 customers in 64 countries, with 80% of our clientele comprising listed companies in major markets such as Europe, North America, and the Middle East. Our Recognition: DiliTrust has been at the forefront of Legal Tech innovation, being the first Legal Tech with AI features since 2022. The company is renowned for providing a positive and entrepreneurial work environment. We are honored to have received the "Happy at Work" and "Tech at Work" labels every year since 2019.

The Role

Lini is DiliTrust's proprietary AI engine, powering Ask Lini, Risk Detector, Document Summarization, Minute Generation, and every AI capability across the suite. We are building a dedicated squad around it and are looking for a strong Software / Product Engineer, with a focus our platform architecture, to help us bring it to the next level. As a Software Engineer work at the intersection of AI, product, and engineering. You will contribute to building, improving, and scaling the features that make Lini a reliable and powerful AI layer across the entire DiliTrust suite. We are looking for an engineer who writes clean, production-ready code and is comfortable taking ownership of features end-to-end, from technical design to deployment. We also care about how you think about AI: whether you bring genuine curiosity to the product, and whether you can translate a model capability into a great user experience. Missions Write specifications as the durable asset of the project. Executable acceptance criteria, and — crucially — the non-functional requirements that specs almost never carry: data classification, endpoint × role authorization matrix, volumetry assumptions, latency and throughput budgets. These are the requirements whose later correction is superlinear, so they get decided before generation, not after. Freeze the contracts before any fan-out. API schemas, types, module boundaries, invariants. Agents do not negotiate an interface in the hallway: each will make a plausible and incompatible assumption, discovered at integration. Pilot coding agents with a tight brief and a bounded context — narrow tasks, defined input and output artifacts, explicit stop conditions and budgets, full traceability of what produced each change. Keep producer and verifier separate. You do not sign off alone on generation you piloted, and you act as independent verifier on your peers' slices — with an adversarial brief ("find what breaks against this spec"), never "confirm this looks fine". Build the asymmetric gates that make the slice safe at volume — expensive to satisfy, cheap to check: property tests, contract tests, query-count and allocation budgets, execution-plan checks, policy-as-code, backward-compatibility proofs. Written before the implementation exists, so the agent closes the feedback loop itself without consuming human attention. Use code reading as a calibration instrument, not as a gate. Sample deliberately, by risk zone, to measure the real defect rate of the generator-plus-gates pair and to keep the team's mental model of its own system alive. Keep work-in-progress low. Capped PR size, thin vertical slices, trunk-based with very short branches, feature flags over long-lived branches, CI as the agent's first task rather than its last. More features in flight does not mean faster delivery when the bottleneck is verification. Design for blast radius. Reversibility, progressive delivery, structured logs, traces, metrics and instrumentation generated as a matter of course. On many paths, detecting in five minutes beats three days of review that prevents nothing.

Requirements

Being based in France with full working rights. Fluent in French and English. Experience & Seniority: 8+ years of professional software engineering experience, with a significant portion spent building and operating B2B SaaS platforms in production Proven ownership of features across their full lifecycle — design, delivery, iteration, maintenance — on long-lived, multi-year products Strong background in complex, scalable web architectures, including real modularity: you have seen where coupling stops parallel work dead AI-Native Practice: Demonstrated, sustained use of coding agents in production work — not autocomplete, but delegated implementation with review and accountability A concrete, articulated view of where generated code fails: correlated defects rather than idiosyncratic ones, uniform surface quality that destroys the usual "look here" review signals, plausibility with no author to interrogate Comfort being accountable for code you did not author, and the discipline to refuse a change you cannot explain Verification & Quality: Real fluency in property-based testing, contract testing, and deriving tests from the specification rather than from the code Test-data strategy, including maintaining a volumetrically representative dataset as part of the verification apparatus — not a nice-to-have Instinct for the difference between a rigorous check and a scalable one: does the cost of checking grow with the volume produced? Security & Performance (non-negotiable on this role): Authorization modelling in multi-tenant systems, and why declarative, centrally enforced authorization beats reviewing each endpoint. Generated code reliably checks who