In a world that’s getting used to the idea of an AI agent helping deal with generic workloads, a Palo Alto–based AI startup is building what it calls an “AI co‑scientist” to speed up and improve the way advanced scientific research gets done. The startup, called K‑Dense, has shown potential with its idea and has just been picked for the coveted 2026 Atoms AI Cohort run by Accel and Google’s AI Futures Fund.
Selected as one of five startups selected from more than 4,000 applications worldwide, K‑Dense will now receive capital, cloud compute, and deep product mentorship through the Atoms AI Cohort 2026. The selection in Atoms Cohort 2026 positions this young company as a serious contender for making a mark in the world of artificial intelligence, scientific discovery, and enterprise‑grade research tooling.
The first product to emerge from K-Dense is K-Dense Web, an AI agent constructed to automate literature reviews, experiment design, and generate detailed research proposals. The startup essentially wants its flagship product to sit beside human experts in labs, universities, and R&D teams as a high‑bandwidth collaborator, not just a fancy chatbot.
“K-Dense Web serves professionals across science, finance, engineering, healthcare, and beyond. Our open-source Claude Scientific Skills have been adopted by thousands worldwide,” states the startup.
How K-Dense built an AI co‑scientist
K‑Dense was founded in 2025 and is headquartered in Palo Alto. The startup put its focus on a very specific and demanding use – the working scientist and researcher. Basically, all those who have to engage in technical research and need to be immersed in complex, repetitive knowledge work. With K‑Dense Web, it designed an AI agent that understands natural‑language instructions and turns them into structured research outputs, ready for scientists and researchers to consider.
For example, researchers can give commands like:
– “draft a PhD proposal,”
– “map the literature on X,”
– “find the gaps in Y”
Unlike generic chatbots such as ChatGPT and Gemini, K‑Dense Web is positioned specifically as an AI co‑scientist. Hence, the agent is capable of reading large volumes of scientific literature, synthesising findings, proposing hypotheses, and generating long‑form technical documents – processes that usually take weeks to accomplish by most researchers. K-Dense claims that its product has access to more than 250 databases across science, finance, and clinical domains, with support for over 200 scientific data formats, and the ability to spin up “unlimited tools generated on demand” to handle niche analytical tasks.
As an example, K‑Dense highlights what its AI co-scientist can do in a professional space. The company shows how the generation of a 26‑page PhD proposal on biologically inspired robotic actuators was produced in under 45 minutes, complete with literature synthesis, identified research gaps, and structured sections suitable for academic review.
K-Dense is selling its product on the basis of this killing combo – a blend of speed, depth and domain‑specific capability, which can turn multi-week workflows into a single working session.
The team behind K-Dense
The startup is led by Timothy Kassis, who is a multidisciplinary engineer, scientist, and entrepreneur, leading the development of its AI co-scientist in Palo Alto. Kassis has a PhD in bioengineering and postdoctoral training in biological engineering at MIT, with his career spanning machine learning, bioengineering, and complex systems at the interface of biology, computation, and hardware.
Before K‑Dense, Kassis served as Head of AI and Data Science at Boston‑based Matterworks, rising through roles from machine learning engineer to principal researcher. He had also previously founded Augence Technologies, a startup focused on using AI to enhance the scientific research process. Kassis has also been on technical and scientific advisory roles, teaching as a lead instructor at MIT. His research track covers lymphatic biology, imaging, and microphysiological systems, grounding his AI work in deep domain expertise.
The rest of the founding team at K-Dense brings a mix of scientific and business experience.
– David Zhang, Interim CEO
– Aubrey Brueckner, Head of Operations
– Nicolas Garreau de Loubresse, Head of Business Development, PhD,
– Brian Laffin, PhD, business development consultant
– Darshil Patel and Orion L, Technical staff members
K-Dense Web promises solid help for human scientists
As part of the K‑Dense mission, the company ensures an open foundation that includes its Claude Scientific Skills, which is a collection of over 170 ready-to-use scientific and research skills for any AI Agent supporting the open Agent Skills standard. The skills now include cancer genomics, drug-target binding, molecular dynamics, RNA velocity, geospatial science, time series forecasting, FRED economic data, and more.
K-Dense also states that it wants to keep the human experts firmly in control of all research work, and the company stands by rigorous quality standards – the company says the output is reproducible.
The company is cultivating a wider community through a Slack channel, GitHub organisation, newsletter, and social presences on LinkedIn, X and YouTube. The company is actively encouraging scientists, engineers and analysts to try K‑Dense Web’s “Fast mode,” which is currently marketed as free for a limited time.
K-Dense in Atoms AI Cohort 2026: How it’s going to work
The company’s recent selection for the Atoms AI Cohort 2026 marks a major milestone for the young startup. The programme, run jointly by Accel and Google’s AI Futures Fund, chose just five companies out of 4,000 applicants, which include Dodge AI, LevelPlane, Persistence Labs, Zingroll and K‑Dense.
Each selected startup in the cohort is eligible for co‑investment of up to 2 million dollars from Accel and the AI Futures Fund, alongside up to 350,000 dollars in compute credits on Google Cloud, Gemini, and Google DeepMind infrastructure. The programme will culminate in June, giving K‑Dense structured access to capital, compute, and seasoned operators from both the venture and hyperscaler worlds.
Why K‑Dense has a promising future
With research labs and R&D teams struggling to keep up with fast-expanding literature, limited headcount and rising costs, a startup like K‑Dense offers exactly the kind of assistance that these setups require. The firm is betting that the next productivity leap won’t only come from bigger ‘all-rounder’ AI models but from tightly integrated agents that understand how real research actually gets done.
By embedding its K-Dense Web agent in the workflow of scientists and researchers, the company is positioning the idea of a co-scientist as a practical and necessary tool to speed up research work.