Founding AI/ML Engineer

Lightning Rod Labs trains AI to predict the future. We train LLMs to predict real-world outcomes with scalable self-play on unstructured data. We make it easy for enterprises to train custom models from messy data, no extraction required. We’ve demonstrated that small models (14B & 32B) trained with our method can outperform frontier models on prediction‑market benchmarks using generated data—with no human bottlenecks. Messy data in; accurate predictions out. What You’ll Do You’ll build cutting-edge systems that teach AI real‑world reasoning, at scale. The worlds first platform for Prediction Agents. Create and evolve infrastructure for training and serving prediction agents. Scale. Orchestrate distributed training and inference on large multi-node GPU clusters. Transform messy data. Convert unstructured streams into training signals suitable for learning and evaluation. Learn causes, not just correlations. Design novel objectives, rewards, and calibration methods so models learn causal factors and output well‑calibrated probabilities. Experiment relentlessly. Optimize our training framework, tune specific models, and demonstrate state‑of‑the‑art results on benchmarks. Own delivery end‑to‑end. Ship from data collection to deployment. What We Expect from You Excellent python chops & experience fine-tuning LLMs. Track record of shipping impactful systems. Ownership mindset and comfort operating with autonomy. Proof of skill — GitHub, portfolio, papers, or demos. Scale experience: you’ve worked beyond toy setups—millions of samples and large, distributed GPU clusters. Technical range: rapid analysis in a notebook to robust, scalable data and training pipelines. Tooling: Verl, Hugging Face, PyTorch, pandas, and standard training/data tooling. Data intuition: strong grasp of statistics, probability, data hygiene, and machine‑learning best practices. Cloud fluency: Comfortable across cloud platforms. We work with Modal, GCP, Azure, and AWS just to name a few. What You’ll Get Out of It Cutting edge, for real. Help advance a new way to train LLMs for prediction from unstructured data using RL and self‑play, and build a framework that can eventually teach AI things humans don’t know. Meaningful impact. Work on critical use cases like defense, healthcare, and major institutions. Product ownership. Own major pieces of our stack—from design to shipped systems. Speed & autonomy. High‑trust, fast‑paced environment where you set direction, ship often, and see rapid real‑world impact. Tight‑knit team & company‑wide collaboration. Work closely with a small team and partner across the whole startup (product, engineering, and customer teams) to carry work from research to real‑world impact. Location Remote-first with a growing hub in Brooklyn / NYC (preferred); meaningful overlap with Eastern Time is required. #J-18808-Ljbffr