Built by and for impact researchers, Laude Institute operates at the intersection of academia and industry to catalyze research that reaches people's hands.
Today marks the launch of Slingshots // ONE, the first batch in Laude's initiative to help ambitious research teams ship their work faster. We’re energized by this group of projects advancing the science and practice of AI, and proud to call them part of the Laude community.
Before we get to the full list, a bit of background. Slingshots are no-strings-attached research collaborations and grants designed to help computer science researchers move their work forward. Okay, one string: every recipient pledges to ship their work into the world — through an open-source project, startup, or other artifact.
Each collaboration starts with two simple questions: what are you working on, and what do you need to push it further? We plug in wherever research needs acceleration, from fast-cash grants and compute, to embedded engineering, product support, and one-on-one advising. We accept projects on a rolling basis and deploy resources immediately after selection.
Our goal is simple: get the right resources to the right researchers at the right time.
The idea for Slingshots comes from Laude co-founder Andy Konwinski’s own trajectory helping translate a university research project, Apache Spark, into a $100B+ platform that reshaped how data and AI systems are built at scale. Slingshots extends that same playbook to the next generation of researchers building at the frontier of AI. The program is led internally by Laude Research Partner and Snorkel AI co-founder Braden Hancock, who has also translated academic research into real-world impact, and is supported by our full internal team and our extraordinary network of research advisors and resource partners.
Slingshots // ONE represents the top edge of today’s AI research ecosystem, and includes a mix of teams from leading institutions like Stanford, Berkeley, MIT, CMU, Princeton, UT Austin, Caltech, and Columbia, alongside open-source builders behind some of the most-watched projects in the community. Together, they tell a story about the direction of modern AI: new benchmarks that redefine evaluation, frameworks pushing the limits of agent capabilities, and applied systems bridging frontier science with real-world use.
A special shout-out in this first batch to Terminal-Bench, a Stanford x Laude collaboration led by Mike Merrill and our own founding MoTS Alex Shaw, and advised by Ludwig Schmidt and Andy Konwinski. Our very first Slingshot, this project helped us model out the program and set the precedent. From idea to industry-standard evaluation for AI coding agents in just 126 days, Terminal-Bench shows how becoming a Slingshot can turn research insights into real-world artifacts that create measurable impact.
The projects in Slingshots // ONE are already shaping how the field moves forward. The strength of this first batch sets our baseline high, exactly where we want it to be.
So without further ado, meet Slingshots // ONE:
Evaluation framework and benchmark measuring AI agents' ability to complete tasks in command-line (terminal) environments. Stanford UniversityLaude Institute
A declarative framework for building modular AI software, compiling AI program specs into effective prompts and weights. MIT
Interactive reasoning benchmark built to measure the generalization ability of AI. Humans can do it, AI cannot. Arc Prize Foundation
Benchmark that evaluates LMs on their ability to maintain and update a codebase to compete in multi-round, long-horizon code competitions. Stanford UniversityPrinceton University
A flexible, high-performance reinforcement learning framework for training multi-turn, tool-use language model agents on real-world tasks. UC Berkeley
Training AI agents to learn from experience via reinforcement learning. UC Berkeley
A fast serving framework for large language models and vision language models.
Specializing coding agents to specific/private repos to improve quality and compressing them to run on your laptop. Carnegie Mellon University
Exploring uncertainty quantification in large language models to improve reliability and trustworthiness. Stanford University
Evaluation framework and benchmark for measuring AI agents' ability to optimize code in real-world codebases. University of Texas at AustinCalifornia Institute of Technology
Training LLMs to generate novel research ideas and automatically execute them and learn from the executed outcomes via RL. Stanford University
A general-purpose biomedical AI agent to automate day-to-day research tasks. Stanford University
Developing the first comprehensive benchmark for white-collar AI agents. Columbia University
Pipeline for automatically extracting insights into model behavior from their traces. UC Berkeley
A framework for optimizing systems (agents, prompts, and code) via reflective evolution, and Pareto-aware candidate selection. UC BerkeleyMITDatabricks
Researchers: Tell us what you’re working on and what you need to get it into the world. We accept projects year-round, deploy resources immediately, and announce new batches every few months. Apply here.