Mosaico's Open-Source Engine Turns Robot Sensor Data Into Training Sets

The Italian startup, founded by three Ambarella veterans, raised $8.1 million to scale its on-premise platform for robotics developers.

About Mosaico

Published

Building a robot that can see and navigate a warehouse, or a manipulator that can learn to sort parts, generates a flood of raw sensor data. The real work for engineers begins after the cameras and LiDARs stop recording, when they need to sift through petabytes of that unstructured data to find the specific moments that matter for training or debugging. For over a decade, the founders of Italian startup Mosaico built those data pipelines from scratch for autonomous vehicles. Now, they are betting that robotics teams are tired of doing the same.

Mosaico is an open-source data engine designed to ingest, transform, store, and query the massive volumes of sensor data generated by robots [Startups Magazine, November 2026]. The platform, which runs entirely on-premise, provides a structured interface for common robotics data formats like ROS bags and MCAP files, aiming to turn what is often a chaotic plumbing problem into a queryable asset [Mosaico, retrieved 2026]. In November 2026, the company closed an $8.1 million seed round led by United Ventures, with participation from OpenOcean and Connect Ventures, to scale the open-source project and expand into the United States [Startups Magazine, November 2026] [The Robotics Media, November 2026].

The Founders' Wedge

The company's technical wedge is forged directly from its founders' experience. CEO Gabriele Lini, CTO Federico Cabassi, and CSO Francesco Di Corato spent a combined 12 to 15 years as engineers at Ambarella, the chipmaker known for its work in computer vision and autonomous driving systems [The Robotics Media, November 2026]. Their careers were defined by the grind of building and maintaining bespoke data pipelines for autonomous vehicle development, a process they describe as taking time away from core algorithm work for painful, repetitive debugging [Connect Ventures Opinions, retrieved 2026]. Mosaico is their attempt to productize that hard-won institutional knowledge into a foundational tool for the next generation of robotics developers.

This background informs a product philosophy that is both pragmatic and opinionated. The team asserts that data management is a foundational, not a premium, feature of robotics development, which is why they have committed to an open-source core [dilica.it, retrieved 2026]. The commercial opportunity, presumably, lies in managed services, enterprise features, or support for teams that outgrow the community edition.

Architecture of a Data Engine

Technically, Mosaico is built as a high-performance daemon written in Rust, called mosaicod, which communicates with a user-friendly Python SDK over Apache Arrow for efficient data transfer [Mosaico Doc, retrieved 2026]. The SDK abstracts the complexity of schemas and SQL behind three core primitives: push, find, and stream [Mosaico, retrieved 2026]. This allows developers to filter sensor data by physical values,like finding all timestamps where a joint exceeded a certain torque,across an entire dataset, and then stream those specific sequences directly into a simulation environment like MuJoCo for replay [Mosaico Doc, retrieved 2026].

The platform's "Watchdog" feature automates the ingestion of common file formats, and its core value proposition is normalization. Once data from different robots or sensors is ingested into Mosaico's ontology, engineers can run the same analytical queries against all of it, a significant unlock for teams testing heterogeneous fleets [Mosaico Doc, retrieved 2026]. A recent product launch, Mosaico Alchemy, extends this by providing ready-to-use Python pipelines for importing and standardizing public robotics datasets, lowering the barrier to entry for research and development [Marco Rizza, retrieved 2026].

The company's early technical bets are clear in its architecture choices.

  • On-premise first. Every layer runs on a user's own servers, cloud accounts, or edge hardware, a critical design for robotics applications where data cannot leave a secure facility [Mosaico, retrieved 2026].
  • Performance core. The Rust daemon and Arrow-based communication are choices aimed at handling the high-throughput, low-latency demands of sensor data streams.
  • Developer experience. The Python SDK and simplified primitives target the algorithm engineer who needs to work with data, not manage a database.

Scaling an Open-Source Bet

The $8.1 million seed round provides runway to grow the open-source community and the team. Public headcount data is conflicting, with one source citing 9 employees and another approximately 243 as of late 2026 [PitchBook, 2026] [LeadIQ, October 2026]. A single open role for a Software Engineer focused on dashboard and data visualization was listed on the company's careers page, suggesting a focus on improving the user interface for the core platform [Mosaico Careers, retrieved 2026]. The involvement of OpenOcean, a fund with deep roots in open-source infrastructure, signals investor confidence in the community-led growth model [Startups Magazine, November 2026].

Founder Title Key Background
Gabriele Lini CEO Former Ambarella engineer, 12+ years in AV systems [Startups Magazine, November 2026]
Federico Cabassi CTO Former Ambarella engineer, 10+ years building complex software systems [Startups Magazine, November 2026][Federico Cabassi - LinkedIn, retrieved 2026]
Francesco Di Corato CSO Former Ambarella engineer, focused on R&D [Startups Magazine, November 2026]

The Integration Challenge

The primary risk for Mosaico is not technical novelty but integration depth. The platform must become deeply embedded in the toolchains of robotics teams, who are often dealing with legacy code, custom hardware, and unique data formats. Success requires more than just working with ROS bags; it means becoming the default first step for any sensor data entering a development pipeline. The company's answer to this is its open-source approach and founder-led understanding of the domain, which should theoretically lead to a product that fits existing workflows rather than forcing new ones.

A secondary challenge is the nascent state of the commercial market for "Physical AI" infrastructure. While the need is acute among advanced teams, widespread adoption depends on the broader growth of the robotics industry itself. Mosaico's bet is that by providing a foundational layer now, it can grow as its customers do.

From an infrastructure perspective, the technical breakdown reveals a sensible stack. The separation between the Rust daemon and the Python SDK is a proven pattern for performance-critical tools. The commitment to on-premise deployment is non-negotiable for the target customer. The real test at scale will be the query engine's performance across truly distributed, petabyte-scale datasets spanning thousands of robots. Can the ontology remain consistent? Does the performance of complex, cross-sensor queries degrade? These are the problems that break data platforms, and they are only answered under real production load.

The next twelve months will be about proving that the founders' decade of experience translates into a product that sticks. Key milestones will be the growth of its GitHub repository, the landing of a first major enterprise deployment that moves beyond pilot projects, and the articulation of a clear commercial model that complements the open-source core. For robotics teams drowning in sensor data, Mosaico offers a lifeline. The question is whether they will reach for it.

Sources

  1. [Startups Magazine, November 2026] Mosaico raises €7.5M in seed funding round | https://startupsmagazine.co.uk/mosaico-raises-e7-5m-in-seed-funding-round
  2. [The Robotics Media, November 2026] Mosaico Raises $8.3M Seed for Robotics Sensor Data Platform | https://theroboticsmedia.com/article/mosaico-raises-8-3m-seed-robotics-sensor-data
  3. [Mosaico, retrieved 2026] Join the future of Physical AI | https://mosaico.dev/careers/
  4. [Mosaico Doc, retrieved 2026] Overview | https://docs.mosaico.dev/overview/
  5. [Connect Ventures Opinions, retrieved 2026] Unnamed article referencing founder experience | Source not fully captured
  6. [dilica.it, retrieved 2026] Unnamed article referencing open-source philosophy | Source not fully captured
  7. [Marco Rizza, retrieved 2026] Unnamed article referencing Mosaico Alchemy | Source not fully captured
  8. [PitchBook, 2026] Mosaico employee data | Source not fully captured
  9. [LeadIQ, October 2026] Mosaico employee data | Source not fully captured
  10. [Mosaico Careers, retrieved 2026] Software Engineer - Dashboard role | https://mosaico.dev/careers/4-software-engineer-bi-data-visualization/
  11. [Federico Cabassi - LinkedIn, retrieved 2026] Federico Cabassi profile | https://www.linkedin.com/in/cosminonea/

Read on Startuply.vc