🌟 About You
Do you get excited when an LLM stops being a demo and becomes a reliable workflow that a legal team can trust? Do you like turning vague requests (“summarize this case”, “draft a clause”, “find all exceptions”) into agentic toolchains with clear boundaries, citations, and predictable behavior? Are you hands-on with prompt + context design, tool interfaces, evaluation, and shipping improvements fast—while keeping cost, latency, and legal risk in check?
If you enjoy building AI workflows that feel like product features (not experiments), we’d love to hear from you.
🚀 About Omnilex
Omnilex is a young dynamic AI legal tech startup with its roots at ETH Zurich. Our passionate interdisciplinary team of 14+ people is dedicated to empowering legal professionals in law firms and legal teams by leveraging the power of AI for legal research and answering complex legal questions. We already stand out with handling unique challenges, including our combination of external data, customer-internal data and our own innovative AI-first legal commentaries.
Tasks
🛠️ Your Responsibilities
As an AI Workflows Engineer – Legal Agentic Tools, you will build and ship the agentic workflows that power real legal work: research, drafting, review, and knowledge operations—grounded in sources, permission-aware, and production-safe.
- Agentic workflow design: Build multi-step workflows (plan → retrieve → reason → verify → output) with clear state, guardrails, and tool boundaries.
- Tooling & integrations: Define and implement tools the agent can safely use (search, document fetch, citation extraction, structured outputs, internal APIs), including robust schemas and failure handling.
- Context engineering: Design how the agent builds context (chunking, routing, citations, memory constraints), and enforce “grounded answers only” behaviors.
- Reliability patterns: Add pragmatic controls: timeouts, retries, circuit breakers, fallback routes, confidence/uncertainty behaviors, and “no citation → no claim.”
- Evaluation & iteration: Build lightweight eval sets for workflows (“must-not-fail” cases), run fast error analysis, and ship improvements weekly.
- Quality signals: Create observable metrics for agent performance (tool success rates, citation coverage, refusal correctness, latency/cost, user edits/acceptance).
- Prompt & model strategy: Decide when to use which model (and why), manage prompts as versioned artifacts, and optimize for cost/latency without losing quality.
- Security & privacy awareness: Ensure workflows respect permissions, avoid leaking sensitive data through prompts/logs, and behave safely under prompt injection attempts.
- Collaboration: Work closely with legal experts and customer-facing teams to translate high-value tasks into reusable workflow primitives and playbooks.
Requirements
✅ Minimum qualifications
- Proven hands-on experience building LLM workflows that went from prototype to reliable production use.
- Strong engineering skills in TypeScript/Node.js (our core stack); ability to write clean, testable code.
- Experience with at least one agent/workflow framework (e.g., LangChain, LangGraph, DSPy) or a strong track record building equivalent custom orchestration.
- Practical understanding of retrieval + grounding (RAG), citations/provenance, and structured output constraints.
- Strong debugging instincts: you can trace failures across prompts, tools, data, and model behavior—and fix them systematically.
- Ownership mindset, clear communication, and a bias for shipping while managing risk.
- Proficiency in English.
- Availability full-time. On-site in Zurich at least two days per week (hybrid).
🎯 Preferred qualifications
- Experience building tool-using agents in high-stakes settings (legal, finance, healthcare) where traceability matters.
- Familiarity with permission-aware retrieval and multi-tenant data boundaries.
- Experience with evaluation pipelines (human-in-the-loop labeling, lightweight dashboards, AI-as-judge used carefully).
- Familiarity with our stack: Azure / NestJS / Next.js, plus search components (Azure AI Search, pgvector/PostgreSQL, OpenSearch/Elasticsearch).
- Experience operating LLM workflows in production (monitoring, incident response, rollbacks, prompt/version management).
- Working proficiency in German (many sources are in German and we speak with German-speaking customers).
- You have a Swiss work permit or EU/EFTA citizenship.
- Knowledge and experience with legal systems, in particular Switzerland, Germany, USA 🧑⚖️
Benefits
🤝 Benefits
- Product-level impact: you’ll build the workflows users feel immediately—research, drafting, review—grounded, reliable, and fast.
- Autonomy & ownership: own agentic workflow quality end-to-end: tools, prompts, evals, and production metrics.
- Team: work with a sharp, interdisciplinary team at the intersection of AI and law.
- Compensation: CHF 8’000–12’000 per month + ESOP (employee stock options), depending on experience and skills.
We’re excited to hear from candidates who want to build agentic legal tools that are trustworthy in the real world. Apply today by pressing the Apply button.