LLM Engineer - Context Engineering, AI Early Stage Project

X is Alphabet’s moonshot factory with a mission of inventing and launching “moonshot” technologies that could someday make the world a radically better place. We are a diverse group of inventors and entrepreneurs who build and launch technologies that aim to improve the lives of millions, even billions, of people. Our goal: 10x impact on the world’s most intractable problems, not just 10% improvement. We approach projects that have the aspiration and riskiness of research with the speed and ambition of a startup. As an innovation engine, X focuses on repeatedly turning breakthrough-technology ideas into the foundations for large, sustainable businesses.

We are an early-stage project at X working to revolutionize the industrial world by making material transformation intelligent.

Our mission is to reduce the massive waste in material harvesting and processing. This is a growing sector faced with numerous challenges including resource exhaustion, rising energy costs, and a sizable carbon footprint.

We are building a system that combines sensing, multimodal AI, agentic digital twins, and advanced physics-based simulation to automate the continuous optimization of complex industrial processes.

We are looking for an LLM Engineer to build out the cognitive engine of our multi-modal sensemaking platform for the industrial world. In this role, you will solve a massive translation problem by converting the messy, unstructured reality of industrial systems (P&ID diagrams, technical manuals, sensor data, and visual feeds) into structured, queryable Process Knowledge Graphs (PKGs).

You will be architecting Agentic RAG workflows where VLMs (Vision-Language Models) and LLMs reason together to generate digital twins. You will bridge the gap between perception (Computer Vision), real-time sensing, and reasoning (Graph-based logic) to create digital value from complex real-world sources.

• Build and Orchestrate Agent Harnesses: You will design, build, and maintain robust agent harnesses and orchestration frameworks either from scratch or off-the-shelf. You'll own the minute details of setting up, scaling, and debugging these multi-agent workflows.
• Design Asynchronous Systems: You will architect highly performant asynchronous systems to handle real-time data ingestion, tool execution, and parallel agent operations without bottlenecks.
• Optimize Context and Inference: You will apply deep transformer knowledge to optimize the information payload. You'll tackle long context processing, prefill caching, KV cache management, inference throughput, and context compression (token optimization) to make our agents fast, reliable, and cost-effective.
• Create Verifiable Evaluation Infrastructure: You will build automated evaluation systems to score agent outputs against ground truth, driving rapid iteration of context strategies and model improvements.
• Fuse Multi-Modal Perception into Structured Memory: You will engineer pipelines that reconcile noisy signals into unified structured representations, managing external memory so agents can retrieve context the same way they reason about it.

• Bachelor's degree in Computer Science, AI, Engineering, or equivalent practical experience.
• 3+ years of experience in software engineering and applied machine learning (Python, PyTorch, or JAX).
• Experience building custom agent harnesses and orchestration frameworks (state management, tool execution boundaries, sandboxing, etc.). 
• Strong background in building, scaling, and debugging complex asynchronous, event-driven systems.
• Experience with advanced agentic RAG, chunking strategies, vector databases, and managing context windows (compression, summarization routing, and memory hierarchies).
• A "0 to 1" mindset with the ability to thrive in ambiguity and define technical roadmaps.

• Deep intuition about model behaviors including their limitations, failure modes, and hallucinations. Deep, active engagement with current AI research discourse and state-of-the-art developments is strongly preferred.
• Experience implementing self-correcting workflows where model outputs are validated programmatically and failures are fed back as context for retry.
• Strong understanding of evaluation design for generative AI systems, with emphasis on automatic, reproducible, and domain-grounded metrics.
• Experience with VLMs for document or diagram understanding, or multi-model fusion combining vision and language outputs.
• Interest in industrial automation, physics-based simulation, or AI for Science applications.

The US base salary range for this full-time position is $141,000 - $200,000 + bonus + equity + benefits. Within the range, individual pay is determined by work location and additional factors, including job-related skills, experience, and relevant education or training. Your recruiter can share more about the specific salary range for your location during the hiring process.

Please note that the compensation details listed in US role postings reflect the base salary only, and do not include bonus, equity, or benefits.