Research Engineer

Description

Company

Pramaana Labs is a frontier AI lab building the verification layer for AI. Founded in 2025 and headquartered in Palo Alto, we turn complex human knowledge, including tax codes, legal rules, clinical guidelines and security protocols, into a formal representation, so every AI answer can be traced, challenged, and proved. Pramaana's architecture pairs foundation models trained to formalize and reason with a symbolic world model encoded in Lean 4. Pramaana was founded by a team out of Google, DeepMind, and Glean, combining frontier AI researchers and formal methods experts.

Responsibilities

• We’re looking for a Research Engineer to build and improve the systems that turn complex knowledge into formal representations and use them to reason and prove. You’ll work closely with researchers and formal methods engineers on Pramaana’s domain formalizer, autoformalizer, and prover, making these systems more accurate, repeatable, and efficient.
• This is a hands-on role combining software engineering with careful experimentation. Your work could include building model and tool orchestration, improving evaluation, developing data pipelines and training environments, or implementing new representations. You’ll own a focused workstream based on your strengths and the team’s needs, with room to contribute across the research stack.
• You’ll take ideas from an initial experiment to a tested implementation, investigate failures, and measure whether changes improve correctness, coverage, latency, and cost.
• Build next-generation neuro-symbolic AI systems at the intersection of foundation models, formal methods, automated reasoning, and verification.
• Design rigorous evaluation frameworks and research benchmarks to study reasoning capabilities, generalization, robustness, and computational efficiency.
• Develop advanced post-training and data synthesis pipelines, exploring techniques in synthetic data generation, reinforcement learning, and model self-improvement.
• Research and implement approaches to knowledge representation, compositional reasoning, program synthesis, and constraint-based inference for complex real-world domains.
• Drive end-to-end research and engineering, from rapid prototyping and controlled experimentation to ablation studies, performance optimization, and scalable production systems.
• Strong software engineering fundamentals and hands-on experience building large-scale ML systems, research infrastructure, or high-performance training pipelines.
• Experience training or post-training large foundation models, including distributed training, fine-tuning, reinforcement learning, and large-scale experimentation.
• Deep proficiency in Python and modern ML frameworks, such as PyTorch, JAX, DeepSpeed, FSDP, Megatron-LM, or equivalent distributed training and inference infrastructure.
• Experience building agentic AI systems, including tool use, multi-step reasoning, agent orchestration, and evaluation of complex model behaviors.
• Familiarity with modern post-training techniques, including supervised fine-tuning, reinforcement learning, preference optimization, synthetic data generation, and reasoning-focused training.
• Strong research and experimental judgment, with experience designing benchmarks, running ablation studies, analyzing model failures, and translating research ideas into measurable improvements.
• Ability to operate independently at the intersection of research and engineering, rapidly prototyping ambitious ideas and scaling successful experiments into robust systems.
• Especially Strong Candidates May Also Have
• Experience building autoformalization, theorem-proving, or other systems that combine models with symbolic tools.
• Experience with Lean, Coq, Isabelle, SMT solvers, compilers, or program analysis.
• Experience optimizing model inference, parallel workloads, or evaluation pipelines for speed and cost.
• Substantial research code, open-source contributions, or projects that demonstrate strong engineering and experimental work.
• Prior experience in formal verification is not required. We value strong engineering, careful experimentation, and the curiosity to learn how models and formal systems work together.