AI & ML Engineer

You build the intelligence at the core of the platform: the agentic system that turns an office action or an invention disclosure into a filing-ready draft. This is the scarcest and most consequential seat on the squad. You own the orchestrator and sub-skill design, the retrieval that grounds it, and the evaluation that proves it works. You work under the direction of the architect and alongside our AI and ML lead. We value depth in GenAI and agentic systems over years served. Experience in a B2B domain where large language models drive agentic workflow automation is a strong plus.

Responsibilities:

• Design and build the master orchestrator and its specialized sub-skills: routing an input to the right patent-legal skill, defining skill contracts, and composing partial outputs into one coherent deliverable.
• Build retrieval-augmented generation pipelines grounded in the IP data foundation and in a patent and prosecution knowledge graph, so that every generated assertion traces to a located source rather than being produced freehand.
• Make grounding and hallucination control structural, back-checking generated claims and arguments against source disclosures, and enforcing attorney-grade quality control.
• Own the evaluation strategy for output you cannot fully trust: measurement without abundant ground truth, abstention and escalation paths, and confidence thresholds that decide when to route to a human.
• Design human-in-the-loop control so that an autonomous flow pauses at the right attorney checkpoints and remains fully steerable.
• Work with model selection, prompt architecture, structured output, and, where warranted, fine-tuning on the patent corpus.

Requirements:

• Hands-on production experience building agentic or multi-agent large language model systems, covering orchestration, tool calling, retrieval, and RAG, rather than prompt engineering alone.
• Fluency with the current toolkit: frameworks such as LangGraph or CrewAI or their equivalents, vector stores, and observability and evaluation tooling.
• A real grasp of evaluation for generative systems, including how to measure quality when labels are scarce and errors are asymmetric.
• Python depth and comfort owning a system from design through to production.
• An advantage: knowledge graphs or GraphRAG, fine-tuning, and any exposure to legal, compliance, or other high-stakes document domains.