Reinforcement Learning Startups

Coasty provides AI agents that automate healthcare revenue cycle tasks like eligibility verification, prior authorization, claim submission, and denial management by performing repetitive work across EHRs, practice management systems, and payer portals through screen interaction. The platform is designed for healthcare billing teams and operates within customer environments with full audit trails and approval gates for payer-facing submissions.

2 investors

Enact provides post-training data for robotics and physical AI models, running policies on tasks to find failure states and generating targeted demonstration and recovery datasets. It also evaluates model performance in-house to confirm the data resolves those failures.

2 investors

Hillclimb develops training data for recursive self-improvement in AI models, focusing on aggregating human research data and automating reinforcement learning environment creation. The company sells the resulting data to frontier AI laboratories.

2 investors

Magma helps companies that build AI agents package and license their agents' interaction data — traces, corrections, and completed workflows — to AI labs, paying those companies a revenue share. It handles onboarding, data mapping, and packaging so agent logs that would otherwise be discarded can be monetized.

2 investors

Markov Studios captures screen recordings and interaction data from real computer use to create datasets that machine learning models can learn from. The company produces labeled data from real workflows in productivity software and video games.

2 investors

Pathmind applies deep reinforcement learning to simulations and digital twins to help industrial engineers and simulation modelers improve performance in industrial operations, supply chains, and mining. It generates and deploys decision-making policies into operations.

2 investors

Praxis AI collects egocentric video, 3D scans, and multimodal sensor data from industrial and residential work environments through partnerships with large corporations, then sells this real-world training data to AI companies developing physical robotics and other systems.

2 investors

Riften provides a routing layer that sits between a company's applications and large language model APIs, directing each request to the least expensive capable model and using self-hosted open-weight models where possible to reduce inference costs. It also uses the resulting task data to train private, company-owned models on recurring work, serving businesses that run AI-powered applications.

2 investors

Synthetic Sciences develops foundation models trained on research process data to enable autonomous scientific work, combining data capture infrastructure with model training to create AI systems with scientific judgment.

2 investors

ashr provides testing and evaluation infrastructure for AI agents, allowing developers to run tests, catch failures, and debug agent behavior before production deployment. The platform offers real-time tracing, prompt versioning, and automated failure detection for teams building autonomous agents.

1 investor

Cartpole develops simulation environments designed to train advanced machine learning models using reinforcement learning techniques.

1 investor

Snap Poker operates a real-money online poker app with a built-in social feed where players can share hands as interactive reels with voiceover and grow a following. It serves poker players and uses a game-style monetization model rather than charging rake.

1 investor

Flash is a platform that trains custom machine learning models for AI agents using user-provided data. It offers fixed-price training runs with downloadable model weights that can be deployed anywhere, positioned as a cheaper and faster alternative to frontier models for routine inference tasks.

1 investor

Idler develops training environments for reinforcement learning that enable AI models to learn programming through simulation of real-world coding tasks.

1 investor

Kairos provides an applied AI layer that enables enterprises to build specialized agents trained on their proprietary data and workflows. The platform helps critical industries bridge the reliability gap in enterprise AI by embedding operator expertise into customized agents that work within messy real-world business environments.

1 investor

Monte builds post-training and continual learning infrastructure that transforms general AI models into specialized agents adapted to specific organizational workflows. The platform captures training signals from real work, measures performance against actual constraints, and enables reinforcement learning so models improve continuously from experience.

1 investor

Nuntius builds tools to ensure AI models comply with specified constraints and behave reliably in commercial applications.

1 investor

Resonance provides a content repository for marketing campaigns that tailors messaging to individual recipients and automatically archives and recycles effective variations.

1 investor

Sylvian provides tools for organizations to create AI agents tailored to specific business tasks and workflows.

1 investor

Synth provides tools for applied AI research and development, enabling developers to build prototypes, run evaluations, and optimize AI software both locally and on cloud infrastructure. The platform combines open-source components with managed cloud services.

1 investor

Touchy is developing a consumer AI personal assistant designed to perceive a user's physical and digital context and act through their apps, returning minimal information while filtering large amounts of data. The company focuses on agent harnesses, AI alignment, on-device models, and continual learning to build a non-addictive alternative to current AI assistants.

1 investor

Topological develops physics-based foundation models for CAD optimization to help hardware engineering teams accelerate their design and optimization workflows. The platform generates efficient designs for complex mechanical problems by understanding physics, geometry, and manufacturability constraints.

1 investor

Velos replaces outsourced manual work and business process operations with AI-powered automation, handling tasks like data entry, reconciliation, document review, and exception handling for enterprise clients.

1 investor