Applied Compute

Generalists are useful, but it’s not enough to be smart. Advances come from specialists, whether human or machine. To have an edge, agents need specific expertise, within specific companies, built on models trained on specific data. We call this Specific Intelligence. It's what we're building at Applied Compute. We unlock the latent knowledge inside a company, use it to train custom models, and deploy an in-house agent workforce that reports to your team. We work with sophisticated companies that have already captured early gains from general models, like @cognition, @DoorDash, and @mercor_ai. They’re pulling even further ahead with proprietary in-house agents that don’t need to wait for the next public model release. Together, we are building and validating models and agents in days instead of months, achieving state-of-the-art performance on customer evals. Our team has high density and low latency. Our founders all worked on different parts of this problem while they were researchers at OpenAI — @ypatil125 as a key member on the agentic software engineer effort (Codex), @rhythmrg as a core contributor to the first RL-trained reasoning model (o1), and @lindensli as a core contributor on ML systems and infrastructure for RL training. Two-thirds of the team are former founders, and everyone brings a deep technical background, from top AI researchers to Math Olympiad winners. We are backed by $80M in funding from Benchmark, Sequoia, Lux, Elad Gil, Victor Lazarte, Omri Casspi, and others. With their support, we are growing the team, scaling deployments, and bringing to market the first generation of agent workforces built on specific models. In short: 1. We are building Specific Intelligence for specific work at specific companies. 2. That will power in-house agent workforces to support their human bosses. 3. That in turn will unlock AI’s full potential through humanity’s greatest engine of progress: thriving corporations in a free market.

"The craft of engineering the right reward is critical. We work with domain experts at enterprises to build the reward function for the eval. Once you find the right hill to climb, you can throw arbitrary compute at it, but so much craft goes into finding the right hill to climb." Thanks to @nvidia for having us on their "Reinforcement Learning at Scale" panel at GTC. Listen here: nvidia.com/gtc/session-ca…

"Our approach is to build products and conduct research that are in service of accelerated AI deployments. Our platform team builds tools and context primitives that enable faster deployment. Our research team builds frontier systems, including a state-of-the-art RL stack. We then take that research and product and forward-deploy with our customers to help deliver real value." Thanks @foundersysk for having us. Open roles at: jobs.ashbyhq.com/Applied%20Comp…

"Everyone's talking about continual learning. That's entirely where this space is going to go." The Applied Compute platform is architected around that premise: build memory and intuition from fragmented data across your entire org, train reasoning models directly on top of it, and close the loop. A model is just one piece. An agent is where it runs, what tools it has, how permissions and auth are handled, how humans guide and instruct it, and the observability around it all. Every interaction should be treated as a training signal so the system can compound over time. Thanks for having us @tbpn

.@appliedcompute achieving frontier capabilities on APEX Agents with just 2,000 tasks is incredible. Their model can produce complex legal deliverables, redlines, and slide decks. It feels like RL is becoming so powerful that it can quickly saturate any benchmark. The barrier to applying agents to the entire economy is building evals for everything. great work @ypatil125 @rhythmrg @lindensli

We partnered with @mercor_ai to post-train custom models on high-quality expert data from fields like law, investment banking, and consulting. Our latest model ranks #1 on the APEX-Agents leaderboard in corporate law and #4 overall. Domain-specific post-training on high-quality, organization-specific data can systematically close the gap between general AI competence and expert-level reliability, making capable enterprise agents practical and affordable for knowledge-intensive industries. appliedcompute.com/case-studies/m…

Scaling Data leads to SOTA Legal Performance on APEX-Agents @appliedcompute built a custom model (Applied Compute: Small) by post-training GLM 4.7 on nearly 2,000 samples provided by Mercor. It is now top of the APEX-Agents leaderboard in corporate law, with a Pass@1 score of 26.6% and a mean score of 54.8%. Here’s what we learnt 👇

Companies should own their data and intelligence, instead of renting it. That’s the thesis behind Applied Compute. So what does that mean in practice? When we help companies turn their latent knowledge into specialized proprietary agents, each customer deployment is fully isolated and runs inside that customer’s VPC. The point is for customers to own their data and the intelligence derived from it, so nothing needs to leave their environment. They can also self-serve whenever they want. Beyond that, diffusion concern also assumes implementation for one company maps cleanly to another. But what makes enterprises unique is messy context spread across legacy systems, teams, workflows, and years of accumulated data. Done properly, deployment ends up being tailored to those constraints, so you can’t copy it to a competitor and get the same result.

We built a proprietary agent to automate merchant onboarding in collaboration with @AppliedCompute. Menus are inherently messy data structures that are complex to verify correctness. Thus, we co-developed a highly calibrated automated grader to serve as the reward function for RL training. And we saw great results! This was a great example where post-training yielded concrete business outcomes for DoorDash.

DoorDash has learned a lot about shipping impactful AI products through our partnership with @ypatil125 @rhythmrg @lindensli at the @appliedcompute team. We're already seeing additional traction collaborating them in other use cases we hope to share soon.

We partnered with @DoorDash to train a proprietary RL-powered agent that encodes internal QA standards into an automated grader, turning expert judgment into a scalable training signal. The result: a 30% relative reduction in critical menu errors and a production system now live across all US menu traffic. appliedcompute.com/case-studies/d…

Agentic co-workers are on the horizon!! It was awesome to work with @BrendanFoody and the rest of the Mercor team on this! Unlocking the true economic potential of AI necessitates learning from domain experts and how work is done in the real world.

When improving agentic systems on real-world tasks, the quality and trustworthiness of the data is often much more important than the quantity (past reasonable thresholds). It was awesome working with Mercor for precisely that reason – their data captures economically valuable domains like corporate law and is excellent. And it's the same ethos that we bring to our engagements with enterprise customers, where we use tools to first build a high-quality internal dataset with the customer before thinking about hill climbing.

We partnered with @mercor_ai to train knowledge work agents across law, banking, and consulting. Our approach used RL on fewer than 1K expert-authored tasks, full-trajectory logging, and behavioral analysis to pinpoint where models succeeded or failed. What made the difference was closing the feedback loop. Each training run surfaced what the data was actually teaching, where progress stalled, and what to collect next. Fast iteration prevented wasted time and compute on the wrong data. The result is both a better model and a system that improves with use, with learning curves that keep climbing across all three domains.

RL is a powerful mechanism for training company-specific models on their unique work and data. This is what we do at Applied Compute. A key challenge is how to make RL efficient, because we need runs to be fast (delivered in days), cheap (scalable unit economics), and predictable (not just fast, but reliably fast). Here are some takeaways: • Synchronous RL is wasteful with time and compute. • Asynchronous RL is more efficient but introduces staleness, which causes learning instabilities. • Modeling and simulations can help analytically solve for what configuration leads to optimal efficiency. This allows us to rapidly prototype training configurations, without burning expensive compute cycles on trial runs. Two of our co-founders, @rhythmrg and @lindensli, discussed some of this research at @aiDotEngineer recently, with a focus on the following subproblem: what is the highest throughput way to do RL given a maximum staleness and compute budget?

Life update: I left Scale and joined Applied Compute a couple weeks ago. It's been fun👨‍🍳

btw if you wanna see what RLaaS taken seriously looks like, you should check out yesterday's @aidotengineer keynote from Yash's cofounders @rhythmrg and @lindensli was proud to host their first talk since coming out of stealth 3 weeks ago

@bfspector Thank you Ben!

@appliedcompute congrats!!

Generalists are useful, but it’s not enough to be smart. Advances come from specialists, whether human or machine. To have an edge, agents need specific expertise, within specific companies, built on models trained on specific data. We call this Specific Intelligence. It's what we're building at Applied Compute. We unlock the latent knowledge inside a company, use it to train custom models, and deploy an in-house agent workforce that reports to your team. We work with sophisticated companies that have already captured early gains from general models, like @cognition, @DoorDash, and @mercor_ai. They’re pulling even further ahead with proprietary in-house agents that don’t need to wait for the next public model release. Together, we are building and validating models and agents in days instead of months, achieving state-of-the-art performance on customer evals. Our team has high density and low latency. Our founders all worked on different parts of this problem while they were researchers at OpenAI — @ypatil125 as a key member on the agentic software engineer effort (Codex), @rhythmrg as a core contributor to the first RL-trained reasoning model (o1), and @lindensli as a core contributor on ML systems and infrastructure for RL training. Two-thirds of the team are former founders, and everyone brings a deep technical background, from top AI researchers to Math Olympiad winners. We are backed by $80M in funding from Benchmark, Sequoia, Lux, Elad Gil, Victor Lazarte, Omri Casspi, and others. With their support, we are growing the team, scaling deployments, and bringing to market the first generation of agent workforces built on specific models. In short: 1. We are building Specific Intelligence for specific work at specific companies. 2. That will power in-house agent workforces to support their human bosses. 3. That in turn will unlock AI’s full potential through humanity’s greatest engine of progress: thriving corporations in a free market.

@tejalpatwardhan Thanks Tejal!

@appliedcompute congrats guys!! 🐐

@kevinweil @peterfenton Thank you Kevin!

@appliedcompute @peterfenton Congrats, excited to see what you guys build!

@mavolpi @ypatil125 @rhythmrg @lindensli @michaelzchen5 Thank you for all your support, Mike! Excited to be on the journey together

Delighted to be part of the @appliedcompute journey. Love the thesis of specific intelligence for specific work. Super excites to see where @ypatil125 @rhythmrg @lindensli and @michaelzchen5 will take us!