Cajal

Problem and Solution Report

The Problem: The Reliability Gap in AI and Science Modern scientific discovery and high-stakes engineering (such as quantum computing and finance) increasingly rely on complex computational models and AI. However, standard Large Language Models (LLMs) are probabilistic systems; they are prone to 'hallucinations' and cannot provide absolute guarantees of correctness. In fields like cryptography, aerospace, or quantum algorithm design, a single mathematical error can lead to catastrophic failure or the loss of millions of dollars. There is a critical need for a system that combines the creative discovery power of AI with the absolute rigor of formal mathematical verification.

The Solution: Cajal's 'Tau' and Lean Integration Cajal solves this problem by 'massively scaling formal verification.' Their core product is a multi-agent system named 'Tau,' described as a 'multi-agent math genius.' Tau is designed to collaborate autonomously to discover and verify mathematical proofs. Unlike standard AI that merely generates text, Tau's outputs are grounded in Lean—a formal proof assistant and programming language. Lean allows for the formal verification of any mathematical statement, ensuring that every result produced by Cajal's system is machine-verified by Lean's type-checking kernel.

Value Proposition and Impact The value proposition of Cajal's approach is the 'bridging of natural language reasoning with formal verification at scale.' By using a multi-agent architecture, they can explore mathematical spaces more efficiently than human mathematicians while maintaining a 'ground truth' through machine verification. This provides 'proven correctness and speedup' for quantum computing and 'provable guarantees' for quantitative finance. The ultimate goal is to accelerate scientific discovery by providing researchers with tools they can trust implicitly, removing the bottleneck of manual verification in complex domains like robotics, biology, and aerospace.

Market and Competitors Report

Market Landscape Cajal operates in the emerging 'AI for Science' and 'Automated Theorem Proving' (ATP) markets. The target audience includes R&D departments in quantum computing, quantitative hedge funds, and engineering firms in high-reliability sectors like aerospace and cryptography. The market is currently shifting from purely academic research into commercial applications, driven by the increasing complexity of software and the need for 'provably correct' systems in the age of AI.

Competitive Landscape Cajal faces competition from three primary groups:

1. Academic Research Labs: Institutions like Princeton University are actively developing AI-powered theorem provers (e.g., models tested on PutnamBench). These groups often release open-source tools that set the state-of-the-art benchmarks for the field.
2. Specialized Formal Verification Firms: Companies like Certora and CertiK focus on formal verification for smart contracts, while firms like Trail of Bits and Runtime Verification provide high-end formal methods consulting for aerospace and defense. These firms are established but often rely on more manual or semi-automated processes compared to Cajal's multi-agent AI approach.
3. AI Research Giants: Organizations like OpenAI and DeepMind have also explored automated theorem proving (e.g., DeepMind's work on AlphaGeometry). While these are not direct 'competitors' in the startup sense, their underlying research often powers the tools that startups like Cajal build upon.

Competitive Advantages Cajal's primary advantage lies in its specialized focus on the Lean ecosystem and its 'Tau' multi-agent architecture. By specifically targeting quantum computing and finance, they are building domain-specific expertise that general-purpose AI labs may lack. Furthermore, their integration with Lean's kernel provides a 'hard' guarantee of correctness that distinguishes them from standard AI coding assistants (like GitHub Copilot) which can suggest code but cannot prove its mathematical properties. As a YC-backed company, they also benefit from a network of early adopters and investors in the Silicon Valley ecosystem.

Founders and Professional Background

Cajal was co-founded by Luke Johnston and Pedro Nobre. The team is currently composed of these two active founders, who established the company in 2025. The venture is part of the Y Combinator Winter 2026 batch, indicating a very early-stage but high-potential trajectory within the Silicon Valley startup ecosystem.

Luke Johnston serves as a primary technical lead for the company. His professional background is deeply rooted in machine learning and neuroscience, having conducted research at prestigious institutions including the University of Cambridge, University of Oxford, and University College London (UCL). This academic pedigree in complex computational systems and neural architectures provides the foundational expertise required to build the 'Tau' multi-agent system that Cajal utilizes for formal verification.

Pedro Nobre is the other co-founder, bringing a background in software development and startup acceleration. He has been recognized with several honors, including a 3rd place finish in the Biolynx Open Innovation Competition in March 2025 and being a finalist in the Green Algorithms Hackathon organized by the Spanish Government in October 2024. Additionally, he participated in the NYU Startup Accelerator Program in late 2023, suggesting a strong foundation in entrepreneurial operations and product development.

Together, the founders combine high-level academic research in AI and neuroscience with practical experience in hackathons and startup accelerators. This blend of deep technical knowledge and agile development is central to their mission of scaling formal verification through AI-driven 'superhuman' mathematicians.