Senior Backend Engineer — Data Platform & AI Agents

About the Role

Jus Mundi is building the AI layer for international law and arbitration — turning a vast, multilingual corpus of case law, treaties, and legal analysis into answers practitioners can trust. We are now expanding beyond arbitration into new legal verticals, and each one potentially adds millions of documents to that corpus. To support that, we are building a new Data Platform: an autonomous, agent-driven system that discovers, acquires, processes and enriches legal documents end to end, and becomes the company-wide source of truth that Search, AI and Product all consume. It replaces our legacy ingestion paths rather than patching them. We are looking for a Senior Backend Engineer with hands-on AI experience to build it with us. You will own the services behind the platform: the agents that acquire documents from hundreds of heterogeneous sources, the transactional catalog and API that everything reads from, and the processing layer that turns raw files into searchable, well-described documents. This is a hands-on senior role with high ownership and real architectural latitude. Backend depth is the core of the role; you should be comfortable working with LLMs and agent orchestration as a first-class part of the system, but this is not a modelling or research position.

The Team

You will join the Data team within EPD (Engineering, Product and Design) — a small, senior team made up of an Engineering Manager, a backend/full-stack engineer, and an AI engineer reinforcing us over the coming months. We work in close collaboration with the Legal Content team, who own what “correct” means for a legal document, and with the Search, AI and Product squads who consume what the platform produces.

Your mission at Jus Mundi

Agentic ingestion (core of this role): Build the agents that acquire documents from hundreds of heterogeneous sources — the orchestration loop, tool calling, memory, and the guarantees that make a sweep provably complete rather than approximately complete. Backend services and data model (core of this role): Own the transactional catalog and the API in front of it — the single door to the platform’s data. Design a schema that holds heterogeneous, evolving legal documents with provenance and confidence on every field, and that absorbs new contributors without a migration. Document processing: Extraction, OCR and metadata over a messy multilingual corpus. Derived data stays versioned and regenerable, so a better model means a re-run rather than a re-acquisition. Quality without manual gates: We deliberately have no human approval step before a document becomes searchable, so quality has to be measured. Test sets, thresholds, confidence and regression tracking are how a component earns its way into production. We will develop Human-In-the-Loop architecture Run what you build: Deploy, instrument and operate your own services. Metrics, logs and traces that let anyone reconstruct a single document’s whole history from one identifier. Leverage across squads: Search, AI and Product all build on this platform. You design the stable interfaces they consume, write the RFCs that get decisions validated, and help us grow the team. Scope will move as the platform does — this is an early-stage system, and we would rather you help shape the roadmap than inherit it.

Backend Engineering Depth

5+ years shipping and operating production backend or data services, with strong Python . Async I/O, API design, testing discipline, and a track record of owning services in production rather than handing them over.

Agentic & LLMSystems

Hands-on experience building agent orchestration in production — tool/function calling, memory and state management, error recovery, bounded context, and deterministic contracts wrapped around a non-deterministic core. Custom architecture or a framework (LangGraph, LangChain, LlamaIndex, or similar) are equally welcome; what matters is that you can point at something you shipped and had to keep working.

Data Modelling

Postgres in depth — schema design for heterogeneous and evolving data, jsonb versus typed columns, indexing, full-text search, migrations under load, and a real feel for transactional correctness.

Systems Judgment

Queues or streams, idempotency, at-least-once semantics, atomic writes, object storage treated as immutable truth. You design for retry, reproducibility and graceful degradation rather than assuming the happy path.

Document & Data Processing

PDF and HTML text extraction, OCR, and structured metadata extraction with LLMs — plus realistic judgement about where quality breaks and what that costs downstream. Multilingual corpora a strong plus.

Ownership & DevOps Mentality

Containers, CI/CD, Kubernetes or equivalent. You build it, you run it, you own it — including when it pages.

Senior Behaviours

Turns ambiguity into a plan and a document others can act on; makes explicit trade-offs rather than reaching for the perfect solution; mentors peers; takes accountability for outcomes. Fluent English (our working language, oral and written).

Nice to Have

Web crawling and scraping at scale — pagination traps, rendering, rate limits, source drift, dedup. We are taking an agentic-first approach, so this is a real bonus rather than a requirement. Legal tech, or another domain where being confidently wrong is expensive. Search infrastructure (Elasticsearch / OpenSearch) and multilingual retrieval. Enough TypeScript/JS to contribute to our internal back-office UI. Experience with the MCP ecosystem, or agent tooling you built yourself. Open-source contributions, or a portfolio of systems you can talk through in depth