HybridTwo days on-site per week; relocation assistance to France is available.
Software EngineeringBackend Engineering
About the job
Requirements
The candidate must have built and run backend services in Python in production, be able to deliver a well-scoped feature independently, investigate production failures, reason about data consistency, write tests, and design APIs.
The candidate must have solid SQL and Postgres skills, including table design, indexes, migrations, and transactions.
The candidate must have production experience with Docker, CI/CD, logs, and metrics.
The candidate must be fluent in written and spoken English.
Responsibilities
Own delivery of the assigned area from design through production, participating in architecture decisions alongside the Engineering Manager and senior engineers.
Add capabilities to end-to-end pipeline agents, improve their reliability, and onboard new sources; take a new source or jurisdiction from its first crawl to a searchable document within the first six months, with a senior engineer reviewing the design.
Build and improve checks that tie extracted values to source-text quotes, measure quality on evaluation sets, and route doubtful cases to human reviewers in collaboration with the Legal Content team.
Build and improve retry-safe workflows that avoid duplicate documents, minimize repeated model calls, and make processing costs observable.
Add endpoints and write migrations for the FastAPI service and Postgres catalog, maintaining complete and consistent lineage for documents, versions, and the sources of extracted values.
Deploy services, monitor logs, metrics, and traces, fix production issues, and track the cost of processing documents.
Ship a fix or improvement to an existing agent within the first two weeks.
Add a new metadata field end to end, covering extraction, source checks, evaluation, and the API.
Investigate a production failure and make that class of failure retry-safe.
Work with legal experts to bring a new data source into production and own the task while working with a senior engineer.
Desired Qualifications
The candidate has built a real project using an LLM, such as tool calling, structured output, or an agent loop, and has addressed model mistakes through validation, retries, or evaluation; any framework or custom code is acceptable.
Experience with web crawling, including rate limits, pagination, and changing websites.
Experience with document processing, including PDF and HTML extraction, OCR, and messy or multilingual text.
Experience with queues and background jobs, such as Celery, RabbitMQ, or similar technologies.
Experience with search and embeddings, including Elasticsearch, OpenSearch, or vector search.
Experience with LLM observability tools such as Langfuse or similar.
Some TypeScript and Vue experience.
Experience in legal technology or another field where confidently incorrect answers are costly.
Join Jus Mundi – To Power Global Justice 🚀⚖️
Founded in 2019 and recognized as a mission-led company, Jus Mundi is pioneering legal technology to power global justice through artificial intelligence. 🌍
Headquartered in Paris, with additional offices in New York, London, and Singapore, Jus Mundi serves over 150,000 users, including top law firms, multinational corporations, government bodies, and academic institutions in more than 90 countries. By combining legal expertise with cutting-edge AI technology, we provide global legal intelligence, data-driven arbitration professional selection, and business development services.
What Makes Jus Mundi the Place to Be 🌟
🧬Our Values: With core values asListen (Solve for the Legal Community), Iterate (Start Sharp, Stop Smart), Commit (Step In, Own the Outcome), Belong (Build together through diversity), Uphold (Aim for Justice with Integrity), we cultivate a dynamic environment that encourages collaboration, innovation, and continuous growth.
🚀 Legal Is the Next Frontier for GenAI: Legal is one of the top industries ready to be transformed by GenAI. At Jus Mundi, we’re building transformative AI where it has the power to make the most impact.
🌐 Proven Product Market Fit: Jus Mundi has secured product-market fit in global markets since launch, partnering with top-tier law firms, government bodies, and major corporations, including: Freshfields, Quinn Emanuel, Wong Partnership, Total Energies, Canada, UK, and Japan.
💸 Strategic Investment: In 2024, we raised over $20 million in Series B funding from prominent strategic international investors, including Acton Capital, True Global Ventures, C4 Ventures, and FJ Labs, fueling our growth and innovation.
🤖 Legal Meets Tech: Our engineers and former lawyers collaborate closely with large language models (LLMs) to pioneer the future of generative AI and reshape professional services.
🌟 Exceptional Team: Work with talented professionals from industry leaders, including engineers, legal experts, and AI specialists from companies like Google and Amazon who are pushing the boundaries of legal tech.
📌 About the Role
LLMs are confidently wrong. In international law and arbitration, a wrong date, party name or tribunal in front of a lawyer is not an acceptable error.
Jus Mundi is building the AI layer for international law. We are now expanding into new areas of law, and each one brings new sources with thousands of documents in different languages.
To handle that, we built a Data Platform: a pipeline of AI agents that finds legal documents on court and institution websites, downloads them, reads them and describes them, and has to prove every value it extracts. What it produces feeds our Search and Jus AI products, which lawyers use to prepare international disputes.
It already runs end to end. Several design questions are still open, and the engineer who joins now will help answer them alongside the EM and senior engineers:
How should a workflow recover from a partial failure without repeating expensive model calls?
How should evaluation gates balance automation, human review, quality and throughput?
How should we version documents, prompts, models and extracted evidence so reprocessing stays safe?
🌟Your mission at Jus Mundi
You own the delivery of your area, from design to production. Architecture decisions are made together with the Engineering Manager and senior engineers, and you take part in them.
Agents, end to end. Each pipeline step is an agent: crawl a court website, download, read, classify, extract metadata, summarize, index for search. You add capabilities, make them reliable, and onboard new sources. Within your first six months, you take a new source or jurisdiction from first crawl to searchable document, with a senior engineer reviewing the design.
Proof, not confidence. Every extracted value is tied to a quote from the source text, so we can always show where it came from. A quote alone does not prove the value was read correctly, so we also measure quality on evaluation sets and route doubtful cases to a human reviewer. You build and improve these checks, with lawyers from our Legal Content team defining what "correct" means.
Cost and failure as design inputs. Every document has a processing cost, and every model call is paid for. Jobs fail and get retried. You build and improve retry-safe workflows that never create duplicate documents, keep repeated model calls to a minimum, and make processing cost observable.
Lineage for every value. A FastAPI service and a Postgres catalog hold every document, its versions, and where each value came from: which source, which model, which prompt, which reviewer. You add endpoints, write migrations and keep that lineage complete and consistent.
What you ship, you run. You deploy, watch logs, metrics and traces, fix what breaks, and know what each document costs to process.
Your first 90 days
Ship a fix or improvement to an existing agent in your first two weeks.
Add a new metadata field end to end: extraction, check against the source, evaluation, API.
Investigate a production failure and make that class of failure retry-safe.
Work with our legal experts to bring a new data source into production and own the task while working with senior engineer.
Stack: Python, FastAPI, Pydantic AI, RabbitMQ, Postgres, Elasticsearch, Langfuse, Kubernetes, Vue for the back office. Nobody expects you to know all of it.
🙌 The Team
You join the Data team in EPD (Engineering, Product and Design): an Engineering Manager, a backend/full-stack engineer and an AI engineer. Four people means your work is visible and your input on design is heard. You get regular code review and pairing, and a clear area to own from day one.
You work daily with the Legal Content team (lawyers who set the quality bar) and with the Search, AI and Product squads who build on what the platform produces.
⚙️ Preferred Experience and Skills
Must have
You have built and run backend services in Python in production. You can deliver a well-scoped feature on your own, investigate a production failure, and reason about data consistency. You write tests and design APIs.
Solid SQL and Postgres: table design, indexes, migrations, transactions.
Production habits: Docker, CI/CD, logs and metrics.
Fluent English, written and spoken.
Strongly preferred
You have built something real with an LLM, at work or in a side project you can show us: tool calling, structured output, an agent loop. You have dealt with model mistakes through validation, retries or evaluation. Any framework or your own code is fine.
If you are a strong backend engineer who has only started experimenting with LLMs and wants to make agents your job, apply. We value solid engineering over a long AI résumé.
Nice to have
Web crawling (rate limits, pagination, sites that change)
Document processing: PDF and HTML extraction, OCR, messy or multilingual text
Queues and background jobs (Celery, RabbitMQ or similar)
Search and embeddings (Elasticsearch, OpenSearch, vector search)
LLM observability (Langfuse or similar)
Some TypeScript and Vue
Legal tech, or any field where a confident wrong answer is costly
How you work: you ask questions early, write down what you decided, take feedback well, and care whether the result is actually right.
🔄 Interview process
We’re proud of our multicultural team representing 38 nationalities! To ensure smooth communication, please apply in English (CV + application form).
Intro call with Laurine, Talent Manager (45 min)
Technical session (1h30): a realistic pipeline problem where AI tools are allowed. We look at your judgement: how you check the output, handle retries and cost, and model the data. Then a deep dive into something you built. No algorithm puzzles.
Product interview with an Engineering Manager and a Product Manager, remote (45 min)
Meet the team at the office, then lunch or afterwork
Final call with Christophe, our CTO (30 min)
Please note that, due to the nature of the sector we operate in, the final steps of our recruitment process include reference calls and a background check.
Need More Reasons to Join Us? ✨
💻 Enjoy a hybrid working organization that combines remote and on-site work (2 days).
🏖 Get 5 weeks of vacation each year, and take them whenever it suits you.
🩺 Receive 100% coverage of private health insurance for yourself and your children.
🍴 Enjoy 60% company-funded restaurant vouchers for your weekday meals (Swile).
😍 Participate in company-wide summits called JM Fest to connect and inspire.
🤸♂️Stay active with a co-financed ClassPass membership by accessing a variety of fitness and wellness activities.
🌍 Enjoy the flexibility to work from anywhere across the globe for up to 8 weeks per year.
✈️ Receive relocation assistance to help ease your transition to France.
💙 Last but not least: Experience a commitment to building a warm, inclusive, and diverse environment.
At Jus Mundi, we believe that diversity, equity, and inclusion are key to our success. Confidence can sometimes hold us back from applying for a job, but we’ll let you in on a secret: there’s no such thing as a ‘perfect’ candidate. We celebrate diverse backgrounds, experiences, and identities, and we encourage you to apply regardless of how you identify. Whether you’re early in your career or a seasoned professional, your unique perspective and skills are what matter most. Join us in building a workplace where everyone feels valued, empowered, and excited to contribute their best work every day.