LiveEO leverages satellite imagery and artificial intelligence to provide actionable insights that drive decision-making across various industries. For example, we detect trees and their height in high-resolution imagery to help protect power grids, identify dangerous construction work around critical infrastructure, and ensure compliance with deforestation regulations.

We are looking for a motivated ML Engineer (NLP) to join LiveEO’s AI team. In this role you will contribute to the development of our geospatial AI solutions by helping Large Language Models understand geospatial data.

For this position we are looking for someone with experience in fine-tuning LLMs, benchmarking LLMs and validating their outputs. You will work closely with other members of the AI team to build, validate and deploy systems that can understand and reason over large volumes of geospatial data, enabling users to query and understand complex information through text.

• Implement cutting-edge NLP techniques. Stay up to date with the latest research in LLMs and NLP. Rapidly prototype and integrate promising techniques into production-ready solutions tailored for geospatial data.
• Fine-tune open-source LLMs. Adapt and extend existing open-source models to LiveEO’s domain-specific needs. Manage data preparation, training, evaluation, and efficient deployment of fine-tuned models.
• Develop evaluation frameworks for LLMs. Automate evaluation with benchmark pipelines. Define test specifications, and track model performance over time.
• Benchmark different LLMs. Compare open-source LLMs across a range of geospatial tasks. Measure and document strengths, weaknesses, and trade-offs, including performance, accuracy, latency, and cost.
• Deploy LLMs in cloud environments. Build and maintain scalable infrastructure to serve models. Optimize for cost, latency, and reliability. Use orchestration frameworks to manage distributed training and inference workloads.
• Identify and prevent hallucinations. Research and implement grounding, consistency checks, schema validation, and other techniques to reduce hallucinations and improve factual accuracy in responses.

• Excellent Python programming skills with a strong emphasis on writing clean, efficient, and maintainable code.
• Proven track record of applying LLMs to natural language tasks.
• Has experience evaluating and benchmarking LLMs, including creating ad-hoc benchmarks and focused datasets to probe specific behaviors, and translating research questions into small, reliable test sets with clear metrics.
• Experience fine-tuning LLMs
• Open mind, respectful attitude towards others and their work and good communication skills are a must.
• Understanding of geospatial data is a plus.
• Experience with synthetic data generation: designing prompts/specs, setting acceptance criteria/quality filters, deduplicating and tracking provenance, and validating impact against real data.
• Experience with post-training of LLMs, including preference optimization (e.g., DPO/RLHF) and RL with verifiable rewards (RLVR)
• Experience with VLMs is a plus.

• The opportunity to create a product that can improve business processes and lives across the globe;
• Flexible working hours and hybrid work model - we trust our employees to get their work done while maintaining a healthy work-life balance;
• We empower employees to drive their own career development, take initiative and have the freedom to be creative and bold;
• Not an overtime culture - we take care that overtime is done only as a necessity and always offset with time off and rest;
• A collaborative and learning environment - frequent internal workshops, knowledge sharing sessions, journal clubs and hackathons;
• Office located in the centre of Berlin Kreuzberg with free fruit, nuts and drinks;
• Potential to participate in the employee stock option program;
• Urban Sports membership and BVG subsidy, corporate pension program;
• A diverse and vibrant international environment of 30+ different nationalities.

LiveEO is a well funded startup founded in 2018 and based in Berlin. Our primary service is modelling risk to our customers’ assets and infrastructure from vegetation, ground deformation and change detection. We currently have around 160 employees from all over the world with a variety of backgrounds