Participated · Equity · Jul 2026

ZML develops inference-performance software, notably its newly launched LLM inference server ZML/LLMD, which enables a range of open-source large language models to run across Nvidia, AMD, Google TPU, Apple Metal, Intel Arc and other chips. The company’s earlier public project was an open-source inference-focused ML framework released in 2024 and updated in March; ZML/LLMD itself is not open source but launched as a free product to gather usage data. Founder Steeve Morin, formerly VP of engineering at Zenly, leads a tight team of about 20 people based in Paris. ZML positions itself to reduce vendor lock-in and allow enterprises and cloud providers to use mixed chip fleets for cost and energy efficiency. Its cap table includes prominent founders and researchers such as Solomon Hykes, Hugging Face’s Clément Delangue and Julien Chaumond, and Yann LeCun, signaling strong industry support. The company has raised $20 million from a syndicate of VCs to accelerate product development and future releases, while competing with inference-focused startups like Baseten, Inferact, and RadixArk.

Participated · Seed · Jun 2026

Macrodata Labs offers a platform to process robotics demonstrations, iterate on datasets, and scale data workflows so teams don’t have to rebuild their data stack as setups change. The company was founded by Guilherme Penedo and Hynek Kydlíček, who previously worked at Hugging Face on widely used open LLM pre-training datasets including FineWeb and FinePDFs. Macrodata Labs positions itself to address the complexities of physical-world data—large video files, multi-rate sensors, and diverse formats—by providing infrastructure for the robotics data loop. The team announced a $4 million pre-seed round led by Air Street Capital with participation from several VCs and angel investors to pursue this work. The company stated the funds will support development of its platform and scaling of data workflows; no revenue, user, or other operating metrics were disclosed in the announcement.

Participated · Seed · Jun 2026

Concord’s flagship product, Concord Agent, converts creative briefs into live, managed campaigns and handles pacing and optimization against KPIs across multiple media platforms simultaneously. The platform integrates natively with DV360, Meta, YouTube, Amazon DSP, The Trade Desk, TikTok, Google Ads and other major buying platforms. Concord works with advertisers within the WPP and Havas agency footprints and says early users save up to 70% of the time spent on manual campaign execution, optimization, and reporting. The company positions itself as the execution infrastructure for programmatic media buying and plans to expand platform coverage and scale across the US and EMEA. Concord was co-founded by Nathan Venezia (CEO), Nicolas Cosson (CTO), and Antoine Chwalek (Head of Engineering), who bring prior adtech and media exits and leadership experience.

Participated · Equity · Jun 2026

Noa has launched its first mobile application focused on fitness and Pilates, designed as a digital studio emphasizing structured progression, audio coaching, visual design and high-quality content. The experience is conceived primarily for women while remaining open to all users, and the app uses artificial intelligence to personalize programs to users’ goals and levels. Programs are developed with international coaches specializing in Pilates, cardio and strength training. The founders—Guillaume Sztejnberg, Shreyas Rajagopalan and Jason Akakpo—have prior successes: AppTurbo reached €15 million in annual revenue and Green Panda Games amassed over 800 million downloads before being acquired by Ubisoft in 2019. Noa intends to expand beyond fitness into a broader wellness portfolio including nutrition, meditation, sleep and cycle tracking to build a larger premium brand.

Participated · Equity · May 2026

Polsia builds an orchestration platform of AI agents intended to manage core business workflows such as coding, research, cold outreach, paid advertising, and customer support. The founder framed the product as enabling AI to run companies, asserting the approach lowers barriers to company creation by allowing individuals to build with “an idea and a laptop.” Polsia said it ran its own fundraising process on the platform, using the system to manage the data room, investor briefings, and diligence. The company reported it is approaching a $10 million annual run rate while operating with one human and no employees. Polsia named a variety of infrastructure and technology partners, including Anthropic, OpenAI, Stripe, AWS, and others, as part of its ecosystem. The company presented its long-term mission as expanding participation in the AI economy by enabling more autonomous company operations.