We're Applying to the Yandex AI Startup Lab Competition

Happy upcoming New Year, everyone!

We've prepared a special gift under the tree: we've applied to participate in the Yandex AI Startup Lab accelerator, aiming to gain expertise and resources to turn our deep technology into a successful product. Today, we're sharing the details with you.

The application is essentially a detailed written interview about our project: we explain what we're doing, what we've already built, what hypotheses we're testing, how we assess market potential, and so on.

Project Summary

The problem you solve, the essence of the technology

The advancement of AI is creating a new generation of autonomous mobile agents: robots, drones, specialized vehicles, satellites — intelligent and endowed with significant computing resources. However, their group coordination still relies on the traditional IoT approach with centralized cloud infrastructure, creating a single point of failure. If the connection or the cloud fails, the entire system stops, despite the local intelligence of each agent.

We are building Florenetes — the first orchestration platform designed specifically for intelligent mobile agents. It's Kubernetes for devices in motion. Drones, robots, or satellites become nodes of a single distributed cluster.

The essence of the solution:

• Resilience: The distributed control plane operates even when the connection to the cloud is lost.
• Adaptive post-IP Florete Network: Automatically reconfigures topology and routing for moving agents.
• K8s-compatible API: Allows using familiar tools to manage autonomous systems.

Florenetes eliminates cloud dependency, reducing communication and redundancy costs by 30–50% while increasing fault tolerance. We are solving the problem of fragility in autonomous systems tied to traditional clouds and infrastructure communication, paving the way for truly distributed robotics.

A more detailed description can be found on our website: https://rete-labs.com.

What Has Been Done

Research, prototypes, experiments

We have completed the fundamental research phase and developed the full architecture of the solution, consisting of two interconnected parts: the adaptive Florete network and the Florenetes orchestrator. The concept of the Florete network is based on the principles of Recursive InterNetwork Architecture (RINA) and adapted for modern distributed systems. Detailed technical documentation is available for internal use and being prepared for publication. The structure of Florenetes is described on the company website: https://rete-labs.com/technology/.

As part of the implementation, we have created a working software stack in Rust, including:

1. Flor framework for building distributed systems: A ready-to-use asynchronous component framework inspired by the architectures of Fuchsia OS, Android, and the Actor model. It enables the creation and launch of modular applications with dynamic dependency management. This is the foundation for all subsequent components.
2. Florete Forwarding network core: A prototype data plane has been implemented, using secure tunnels based on QUIC and MPLS-like routing independent of IP. Basic APIs for managing services and tunnels are ready, and the recursive mode's operability has been verified. Work is underway on the final modules (link management and forwarding engine).
3. Florete Tunnel Daemon (ftunnel) for orchestration: A working daemon performing functions similar to ztunnel in Istio for Kubernetes. It implements HTTP/TCP/UDP proxying through Florete tunnels and is ready for demonstration scenarios.
4. A benchmark application for testing data plane performance, integrated with Prometheus and Grafana for collecting and visualizing metrics. Used for conducting experiments and regression testing.

Thus, we have not only developed the theory but also created a working prototype of the key components. Finalizing this prototype will allow us to move on to demonstrating the technology's capabilities in specific scenarios.

The entire software stack is being prepared for publication on GitHub (https://github.com/rete-labs/) under the Apache 2.0 license. We are preparing to launch a community of developers and enthusiasts for the new technology.

Hypotheses and Success Metrics

What technical or business assumptions are you testing? How do you measure success?

Technical Hypotheses and Metrics:

1. Recursive architecture is feasible in practice.

   • Status: Verified. The working Rust stack prototype confirms the approach.
   • Success metrics: Throughput (goodput), latency, CPU load.
   • Result: Operates even on weak 32-bit ARMv7 (600 MHz) with predictable overhead.

2. Integration with legacy applications is possible without modification.

   • Status: Verified. The proxy daemon (ftunnel) successfully translates HTTP/TCP traffic.
   • Success metric: Ability to host a standard web application through the Florete network.

3. The network is adaptable to dynamic conditions.

   • Status: In progress. Implementing a Fast Reroute mechanism for rapid route reconstruction.
   • Success metric: Time to switch to a backup route after a link failure ≤ 200 ms.
   • Current result: A design for addressless routing, free from the fundamental problems of IP networks, has been created.

4. The distributed control plane is operational.

   • Proposed solution: A combination of Raft (for stable conditions) and gossip protocols (for maximum instability).
   • Success metric: Distributed service availability ≥ 99.9% at a given network "chaos" level (churn rate).

Business Hypotheses and Metrics:

1. The OpenCore service model will attract paying clients.

   • Status: In progress. Interviews with 2 potential clients revealed demand.
   • Success metrics: Number of interviews conducted (target: 10-20), conversion to paid projects, contract size.

2. Integrators will adopt Florenetes to reduce costs.

   • Status: Not started. Requires a business demo.
   • Success metrics: Number of integrators who tested the demo; number of pilot deployments; estimated CAPEX/OPEX savings for them (target: 30-50%).

3. The project will attract strategic investment.

   • Status: In progress. Preliminary interest has been identified from large IT companies and DeepTech investors.
   • Success metrics: Number of funds that performed due diligence; receiving the first investment offer; valuation at the deal.

Potential Applications

Where can the technology be applied? Specify the segment (B2B/B2C), industries, etc.

B2B Segment (Primary Focus):

1. Industrial Robotics & Automation — wherever coordination of mobile agents is required:
   • Logistics (warehouses, ports, etc.)
   • Agriculture (field work using autonomous agricultural machinery, drones)
   • Resource extraction (quarries, mining, etc.)
   • Construction (road, pipeline, etc.)
   • Security and monitoring (of industrial facilities, power plants, airports, pipelines, etc.)
2. Space & Telecom:
   • Low Earth Orbit (LEO) satellite communication constellations
   • Orbital data centers

B2C Segment (Future Potential):

• Home VPN
• Smart Home
• Personal cloud services/applications (+ App Store for them)

Market Potential

Market size, competitors, uniqueness

Uniqueness: Florenetes is the first platform that combines an adaptive post-IP network (Florete) and a distributed Kubernetes-compatible orchestrator, built specifically for mobile agents. This allows autonomous systems to operate continuously even with a complete loss of cloud connectivity, which is the key differentiator from all analogs.

Competitor Analysis:

1. Traditional Cloud IoT Platforms (MTS IoT HUB, AWS IoT, etc.): Centralized, create a single point of failure, and require constant, costly connectivity.
2. Edge Networking extensions for IoT: Provide only secure connectivity, without built-in orchestration and state management for distributed applications.
3. Edge versions of Kubernetes (K3s, KubeEdge): Retain dependency on a central control plane, not adapted for conditions of constant connection loss and rapid node movement.
4. Industry-specific proprietary solutions** (e.g., Auterion for drones): Closed, vertically integrated ecosystems that do not offer a universal platform, limited to one specific industry and its specifics.
5. Satellite and Telecom solutions (e.g., Starlink): Closed, tied to a specific system, not reusable.

Market Size: We focus on the B2B segment, where our solution reduces CAPEX/OPEX for communication infrastructure by 30–50% by eliminating the single point of failure. The market assessment is based on the target segment of companies using or implementing autonomous mobile agents (robots, unmanned vehicles, drones).

• Total Addressable Market (TAM): The IoT market volume in Russia in 2024 was about ~200 billion RUB with 10-15% annual growth (according to TAdviser).
• Serviceable Addressable Market (SAM): Narrowly targeting logistics, agriculture, and industry (~20-30% of TAM), we get a segment of 40–60 billion RUB (50-120 billion in 3-5 years considering growth).
• Serviceable Obtainable Market (SOM): In the next 3–5 years, through a model of working with integrators and direct consulting projects, we plan to capture a 5% share, which amounts to 2-3 billion RUB at the 2024 level (2.5-6 billion RUB considering growth).

Goals and Motivations for Participation

Why are you joining the accelerator?

We aim to participate in the accelerator to:

• Structure our strategy and business thinking with the help of mentors.
• Create a focused MVP within three months, proving the technology's viability.
• Gain expertise and access to Yandex's resources to accelerate development.
• Build connections with investors and industry partners.
• Win the competition to secure funding and recognition for further growth.

This accelerator could be a key step in turning our deep-tech development into a successful commercial project.

Current Resources

Financing, equipment, partners

Development is currently funded by the team's own resources.

In addition to the core team, we have formed an Expert Council, which includes executives from large IT companies, an expert in DevOps and Kubernetes, and a scientific advisor on wireless communication channels.

Partnerships and negotiations: We are actively negotiating with major IT companies that have shown interest in our technology. These are potential pilot clients or strategic partners for the next stage.

Challenges

What is currently preventing you from completing the project? What problems exist?

The key challenge is the transition from the fundamental R&D development stage to creating a market-oriented product and a sustainable business model. This hinges on two interconnected problem areas:

1. Forming a full-fledged engineering team. Developing the core technology (Florete Network) requires Senior Rust developers with a deep understanding of networks and distributed systems. Hiring such specialists at the current stage is impossible due to limited funding. The existing approach — attracting talent through a future open-source community — is effective for long-term growth but insufficient for meeting aggressive deadlines to create a finished prototype.

2. Product and business model validation. We need to quickly test key technical hypotheses (e.g., about route recovery time and stable operation of the distributed control plane) and package the prototype into a convincing business demo for the first pilots. This requires not only engineering resources but also focused market entry efforts: finding and engaging with integrators, adapting the solution to specific industry use cases, and establishing initial commercial relationships.

Thus, the main obstacle is the lack of resources to simultaneously close critical technical tasks and launch commercialization processes. Solving these challenges directly depends on attracting targeted funding (grants, investments), which would allow us to strengthen the team and focus on market adaptation of the technology.

Discussion

Discuss in Telegram →