Karan Malhotra

Karan Malhotra is an American artificial-intelligence researcher and co-founder of Nous Research, the New York-headquartered open-source AI research company he established in 2023 with Jeffrey Quesnelle, Teknium, and Shivani Mitra. He is the company's Head of Behavior and one of the public faces for its work on the Hermes open-weights language-model family, the Forge agentic platform, and the Psyche decentralized training network built on the Solana blockchain. As of May 2026, he is co-leading Nous through the rollout of Hermes 4 and the production deployment of the Psyche network following the company's $50 million April 2025 Series A led by Paradigm at a reported $1 billion token valuation.

At a glance

• Education: According to public profiles, an Associate of Arts degree from Oxford College of Emory University and undergraduate study in Religion and Philosophy at Emory University.
• Current role: Co-founder and Head of Behavior of Nous Research since 2023.
• Key contributions: Co-founder of Nous Research; collaborator with Teknium on the original Hermes fine-tunes built on synthetic GPT-4 data; co-architect of the Nous strategic shift into DisTrO and Psyche decentralized training; presenter of the WorldSim demonstrations that introduced base-model simulation prompting to a broader audience.
• Public profile: Active in the open-source AI community on X, GitHub, and Discord; podcast appearances on Latent Space, Practical AI, and various crypto and AI shows since 2024.
• X / Twitter: @karan4d
• LinkedIn: karan-malhotra-3b1a781b1

Origins

Public biographical material on Malhotra is comparatively thin: his formative years and early career are documented in interviews and on aggregated business-data profiles, with little press coverage prior to the formation of Nous Research in 2023. According to public profiles aggregated by Bloomberg, ZoomInfo, and others, he holds an Associate of Arts degree from Oxford College of Emory University and pursued undergraduate study in Religion and Philosophy at Emory University in Atlanta, Georgia. Public profiles also list a period as a Machine Learning Researcher at the Stanford Brain Stimulation Lab prior to his open-source AI work; the role and dates have not been independently described in primary press coverage.

In an interview with the Practical AI podcast in February 2024, Malhotra described his entry into language-model work as starting with GPT-2: he discovered a Colab notebook for fine-tuning, and produced experimental tunes including Edgar Allan Poe and H. P. Lovecraft style models. The pattern that recurs in his accounts of the period is hobbyist experimentation evolving into open-source community participation through Discord, GitHub, and X, rather than a credentialed research-lab path.

Career

Before Nous Research, Malhotra contributed to the Open Assistant project under LAION, the German non-profit open-source AI research organization. In the Practical AI interview he described the period as the source of his first non-trivial compute access for fine-tuning experiments, alongside connections to other open-source AI researchers organizing on social platforms.

The pivotal early collaboration was with the pseudonymous AI researcher Teknium, whom Malhotra met on X around the GPT-4 release in March 2023. Teknium had assembled a synthetic-data set built from GPT-4 outputs in a format compatible with Malhotra's own data; the two combined their resources and trained an early model called gpt4-vicuna. The work continued into the original Hermes fine-tune on Meta's Llama base, which became the founding artifact of what would soon be branded Nous Research.

Nous Research formalized in 2023 with Jeffrey Quesnelle as Chief Executive Officer, Malhotra as Head of Behavior, Teknium as Head of Post Training, and Shivani Mitra as a co-founder. The company was incorporated in New York and emerged from the open-source AI Discord community in which the founders had already been collaborating. The Nous-Capybara 7B fine-tune in August 2023, the Hermes line continuing on Llama 2 (7B, 13B, 70B) and later Llama 3 (8B, 70B), and the open-source YaRN long-context paper lead-authored by Quesnelle established the company's brand within the first year. The cumulative download count for Hermes models on Hugging Face exceeded 55 million by 2025.

Malhotra's public role expanded through 2024 as Nous Research moved beyond pure fine-tuning into adjacent open-source AI infrastructure. He presented the WorldSim project on the Latent Space podcast in April 2024, demonstrating a base-model command-line simulation prompting technique that introduced the Nous brand to a broader engineering audience. The 2024 papers on DeMo (Decoupled Momentum Optimization) and the related distributed-training research, co-authored with Diederik P. Kingma, set the methodology base for the company's decentralized training pivot. The Forge agentic-AI development platform launched in late 2024.

The April 2025 $50 million Series A round led by Paradigm at a reported $1 billion token valuation, alongside an additional $15 million from earlier rounds, brought cumulative funding to approximately $65 million and reframed Nous Research as a commercially-funded open-source AI organization. Malhotra has been quoted in primary press coverage on the company's framing of crypto incentive mechanisms as a way to draw idle compute into AI training. The Psyche decentralized training network, built on DisTrO and on the Solana blockchain, launched in early 2025 and produced Hermes 4.3 in 2026 as the first production model post-trained entirely on the network.

Affiliations

• LAION Open Assistant: Contributor, prior to 2023.
• Nous Research: Co-founder and Head of Behavior, 2023 to present.

Notable contributions

Malhotra's public output is concentrated on the founding, public communication, and product direction of Nous Research rather than on individual research authorship at the level of a typical frontier-lab chief scientist. The notable contributions track that emphasis.

• Nous Research co-founding (2023). Established the company alongside Jeffrey Quesnelle, Teknium, and Shivani Mitra with the founding thesis of producing high-quality open-weights fine-tuned language models from the open-source AI Discord community in which the founders had been collaborating.
• Hermes line (2023 onward). Co-architect of the original Hermes fine-tunes on Meta's Llama base, including the Hermes 7B release, with Teknium leading the post-training work. Subsequent generations including Hermes 4 (August 2025) and Hermes 4.3 (2026) have continued the line on Llama 3.1.
• Decentralized training thesis (2024 onward). Public articulation of the Nous Research strategic pivot into decentralized AI training infrastructure, including the DisTrO optimizer and the Solana-based Psyche network.
• WorldSim presentation (March to April 2024). Public demonstration of base-model command-line simulation prompting on Anthropic's Claude 3, popularized through the Latent Space podcast and AGI House SF events.
• DeMo paper (November 2024). Co-author of the Decoupled Momentum Optimization research with Diederik P. Kingma, contributing methodology for distributed training across heterogeneous compute infrastructure.
• Public communication. Spokesperson role for Nous Research in primary press coverage including Fortune, The Block, and VentureBeat, and on podcasts including Practical AI and Latent Space.

Investments and boards

• Nous Research (AI): Co-founder and Head of Behavior, 2023 to present. Privately held; approximately $65 million in cumulative funding through April 2026, including a $5.2 million seed round announced January 2024 led by Distributed Global and OSS Capital, an additional approximately $10 million across earlier rounds, and the $50 million Series A in April 2025 led by Paradigm at a reported $1 billion token valuation.

No public personal angel-investor activity on record outside the Nous Research role in AI, semiconductors, datacenters, software, or energy as of May 2026.

Network

Malhotra's longest-running professional relationships are with his Nous Research co-founders Jeffrey Quesnelle (Chief Executive Officer and Chief Technology Officer; lead author of the YaRN long-context paper), Teknium (Head of Post Training; pseudonymous open-source AI researcher with whom he developed the original Hermes fine-tunes), and Shivani Mitra. The founding team came together through open-source AI collaborations on Discord, GitHub, and X, rather than through a shared employer.

His broader network in the open-source AI ecosystem includes Diederik P. Kingma, with whom he co-authored the DeMo paper on Decoupled Momentum Optimization, and contributors to peer open-source AI organizations including LAION, EleutherAI, and the Hugging Face community where Hermes models are distributed. Malhotra appeared on the Latent Space podcast in April 2024 alongside Joscha Bach of Liquid AI and Rob Haisfield of WebSim.

The Paradigm-led Series A places Malhotra in a working relationship with the crypto-VC ecosystem, including connections to the broader Solana developer community given the Psyche network's on-chain coordination layer. Earlier seed-round participants include Distributed Global, OSS Capital, and Vipul Ved Prakash of Together AI as an individual investor.

Position in the field

As of May 2026, Malhotra occupies a structurally distinctive position among co-founders of commercially-funded open-source AI organizations. Nous Research is among the few open-source AI organizations to have raised a $50 million single round at a $1 billion token valuation, and the only such organization to combine high-quality open-weights fine-tuning with a Solana-based decentralized training network as a core strategic commitment. The token-based valuation framework is unusual among AI organizations and reflects the Paradigm investor base and the Psyche network's on-chain coordination architecture.

The Nous Research origin story differs from peer open-source AI organizations in that the founding team came together through Discord, GitHub, and X rather than through a shared university or research lab. Industry coverage has frequently characterized this Discord-community origin as a structural feature of the company's culture and hiring posture, with public-figure researchers including Teknium operating under pseudonymous handles.

Malhotra's public profile is concentrated in the open-source AI community on X, podcast appearances, and conference talks rather than peer-reviewed academic publication. The DeMo paper with Diederik P. Kingma is the most-cited research artifact on which he is a named co-author. His role at Nous Research, described publicly as Head of Behavior, focuses on model post-training behavior, prompting strategies, and product communication rather than the pure-research authorship typical of frontier-lab chief scientists.

Outlook

Open questions over the next 6 to 18 months:

• Hermes 5 release cadence. The successor to Hermes 4 and 4.3, including capability profile and the base model on which it is built.
• Psyche network production scale. Whether the network scales beyond the Hermes 4.3 post-training run to larger pre-training runs, and the cost and performance economics of distributed Solana-coordinated training relative to centralized Nvidia or Huawei Ascend infrastructure.
• Forge commercialization. Commercial traction of the Forge agentic-AI development platform and the Hermes Agent self-learning runtime.
• Token-issuance details. Specific terms of the token associated with the $1 billion token valuation and the regulatory environment for crypto-aligned AI infrastructure in the United States and internationally.
• Open-weights base-model competition. Pressure on the Hermes fine-tuning thesis as base models from Meta AI / FAIR, Mistral AI, DeepSeek, Alibaba Qwen, and Cohere become increasingly capable and reduce post-training improvement headroom.
• Public profile. Whether Malhotra's role expands further into public communication for Nous Research as the company moves into its first commercially-funded growth phase, or whether Quesnelle and other co-founders take on a larger share of public-facing communication.

Sources

• Karan Malhotra, Nous Research Inc: Profile and Biography. Bloomberg profile listing Malhotra's role at Nous Research.
• Data synthesis for SOTA LLMs with Karan Malhotra, researcher at Nous Research. Practical AI podcast episode #255, February 6, 2024, with the most detailed publicly available account of his pre-Nous career, the Open Assistant period under LAION, and the early collaboration with Teknium.
• WebSim, WorldSim, and The Summer of Simulative AI. Latent Space podcast, April 27, 2024, with Malhotra demonstrating WorldSim alongside Joscha Bach and Rob Haisfield.
• This World Does Not Exist (Latent Space YouTube). YouTube recording of the Latent Space episode.
• Karan Malhotra on Decentralized AI Training & Open Source Models | GLOSAIC. Solo interview with Malhotra on decentralized AI training and open-source model strategy, hosted on the Factorial Computer Club YouTube channel.
• YaRN: Efficient Context Window Extension of Large Language Models. The Quesnelle-led arXiv paper on long-context fine-tuning, with contributions from the broader Nous Research team.
• DeMo: Decoupled Momentum Optimization. The November 2024 arXiv paper co-authored by Malhotra and Diederik P. Kingma on distributed-training methodology.
• DisTrO GitHub repository. Public code and documentation for the Decentralized Training Orchestration optimizer underlying the Psyche network.
• Democratizing AI: The Psyche Network Architecture. Nous Research blog post on the Solana-based Psyche network and its DisTrO-derived training architecture.
• Introducing Hermes 4.3: Local Intelligence Globally Trained. Nous Research blog post announcing Hermes 4.3 as the first production model post-trained entirely on the Psyche network.
• Paradigm leads $50 million Series A round for decentralized AI project Nous Research. The Block coverage of the April 2025 Series A.
• Crypto VC giant Paradigm makes $50 million bet on decentralized AI startup Nous Research at $1 billion token valuation. Fortune coverage of the Series A and the token-valuation framework.
• Nous Research is training an AI model using machines distributed across the internet. VentureBeat coverage of the distributed-training thesis with quotes from Malhotra.
• Nous Research $5.2 million seed financing round announcement. The January 2024 X post announcing the seed round led by Distributed Global and OSS Capital.
• Nous Research Hugging Face organization. Public weights and inference code for the Nous Research model line including the Hermes family.
• Photo: text-mode card generated via scripts/make_lab_card.py --text "Karan Malhotra". No CC-licensed portrait was located through Wikipedia, Wikimedia Commons, or a Nous Research press kit as of May 2026.