AI Engineer

Apilex is an AI-native legal operating system that manages lawyers' entire legal workflows end to end, powered by AI systems purpose-built for each jurisdiction's legal framework. We bring legal research, drafting of pleadings and contracts, document analysis, and case preparation together in a single experience. We currently operate in France, Turkey, and Germany, and we aim to extend the growth we've achieved in our current markets to Brazil, Spain, and Italy. We achieved this growth entirely bootstrapped, without raising any external investment, making us one of the fastest go-to-market success stories among Legal AI startups in the world. Since our founding, we've scaled our team to 120+ people, backed by a proprietary database of millions of court decisions and legal documents that turns days of legal research into seconds. What sets us apart? We don't use off-the-shelf AI. We build domain-specific, closed-circuit AI models with multi-layered verification systems that eliminate hallucinations and deliver answers backed by real citations and legal reasoning. Every feature we ship changes how lawyers work. As a AI Engineer, you will design, build, and operate AI systems used across real legal workflows. You will take ownership of technically challenging projects, contribute to architecture decisions, and work closely with researchers, engineers, and legal domain experts to turn new methods into reliable production systems. Key Responsibilities Design, build, and improve AI systems for complex legal reasoning and long-horizon task execution. Develop agentic systems that can plan, use tools, manage state, recover from errors, and complete end-to-end legal workflows reliably. Build graph-based multi-agent orchestration systems that represent workflows, dependencies, state transitions, and coordination between specialized agents. Design and implement graph engineering solutions across GraphRAG, personalized knowledge bases, and multi-agent orchestration. Build GraphRAG systems that connect structured and unstructured legal knowledge to improve retrieval, reasoning, and source-grounded outputs. Develop personalized knowledge bases that incorporate user, organization, matter, and jurisdiction-specific context while respecting data isolation and access controls. Work on graph schemas, ontologies, entity and relationship extraction, entity resolution, graph storage, and retrieval pipelines for legal knowledge. Optimize agent harnesses, including tool interfaces, context management, memory, orchestration, and execution environments. Contribute to model post-training for advanced legal reasoning, including supervised fine-tuning, preference optimization, and reinforcement learning approaches where appropriate. Develop reward models, verifiers, and automated verification systems for evaluating legal reasoning, citations, intermediate steps, and final outputs. Build scalable pipelines for producing, filtering, and validating high-quality synthetic and human-generated training data. Create evaluation methodologies, domain-expert benchmarks, and regression suites that measure capability, reliability, and behavior across jurisdictions. Improve experimentation and observability systems so that model and agent changes are measurable, reproducible, and safe to deploy. Make sound engineering decisions across model selection, inference infrastructure, latency, cost, reliability, and performance. Participate in architecture reviews, share knowledge with the team, and support other engineers through technical collaboration and code reviews. What We’re Looking For 5+ years of engineering experience building and operating AI or machine learning systems in production. Building AI/ML systems that ran in production — with users, incidents and on-call, not only notebooks. Strong Python. You have designed systems others build on, and you care about boundaries, types and tests. Agentic systems, hands-on. Tool use, state across multi-step tasks, and recovery when a model returns something the schema did not expect. Production RAG. You know why your chunking strategy was wrong the first time and what you did about it. Evaluation design. Benchmark construction, error analysis, model comparison, regression testing. You can tell a real improvement from a lucky sample. Graph-based systems. Knowledge graphs, GraphRAG pipelines, graph databases, or graph-structured agent workflows. Judgement where best practice does not exist yet. Much of this field is three years old; you will have to decide things no blog post has answered. Nice to Have Verifiers, reward models, LLM-as-judge systems, or other automated output assessment, Post-training methods — SFT, preference optimisation, RL, distillation — or long-horizon agents and inference time scaling, Graph databases in anger: Neo4j, Neptune, Memgraph or similar, Ontology design, entity resolution, or permission aware retrieval with hard data isolation, Evaluating reasoning systems in expert or high • Stakes Domains • legal, medical, financial, Legal documents, citation systems, or working alongside domain experts, Open source work, publications or research on agents, evaluation, knowledge graphs or AI reliability. Why join us? Whatever your role, your work here will be visible and meaningful. We're a team that has grown without outside investment and achieved one of the fastest go-to-market successes in the space and we're looking for ambitious people who want to build the future of legal technology on a global stage. This is Apilex. AI for Legal. Join us.