Project Name: InfiniteZero Network — Filecoin/IPFS Integration for Federated AI Training Infrastructure
Proposal Category: Integrations
Individual or Entity Name: Individual — Abraham Nash (on behalf of the InfiniteZero Foundation team)
Proposer: InfiniteZeroFoundation
Project Repo(s)
Filecoin ecosystem affiliations: Abraham Nash completed a Protocol Labs residency. No current active work relationship with Protocol Labs, Filecoin Foundation, or affiliated organisations.
Technical Sponsor: Pending — in conversation with Filecoin ProPGF stewards.
Do you agree to open source all work under MIT/Apache-2 dual-license?: Yes.
Project Summary
InfiniteZero is a trustless, Ethereum-based protocol for a global AI commons — a decentralised network where communities collectively train AI models while raw data never leaves the device. Participants run lightweight validator nodes that help train, aggregate, and validate model updates. Only anonymised, encrypted model weight updates are shared across the network.
Data is the lifeblood of AI. Without it moving freely and reliably through the system, the training pipeline closes down. The InfiniteZero protocol currently uses IPFS for model weight storage and distribution between training rounds — validator nodes push encrypted model updates to IPFS, where they are retrieved by aggregators for the next round. Reliable, persistent, and decentralised storage of these model weights is critical to protocol integrity. Without it, model updates can be lost between rounds, breaking the training pipeline and invalidating contributor work.
This proposal seeks funding to build out the Filecoin integration as a core infrastructure layer for the InfiniteZero protocol — replacing the current centralised IPFS pinning solution with a fully decentralised, trustless, and cryptographically verified alternative. This work demonstrates a novel, real-world application of Filecoin in live federated learning infrastructure, opening a new category of Filecoin usage in the rapidly growing decentralised AI sector.
Impact
Pain points addressed. The InfiniteZero protocol generates encrypted model weight updates at each training round. These updates must persist reliably between rounds — loss of a model update breaks the training pipeline and invalidates that round's contributions. Current IPFS pinning via centralised services introduces a dependency that is inconsistent with the protocol's trustless architecture. Filecoin's cryptographic storage proofs (Proof of Replication and Proof of Spacetime) provide the guarantees the protocol requires.
Benefits of getting this right. A successful integration demonstrates that Filecoin is a viable storage layer for live AI training infrastructure — not just static data archiving. This opens a new category of Filecoin usage in the federated learning and decentralised AI sector. The open source implementation serves as a reference architecture for other projects seeking to build privacy-preserving, decentralised AI training systems on Filecoin/IPFS.
Risks of not getting this right. Without persistent, trustless storage, the InfiniteZero protocol retains a centralised dependency that undermines its core value proposition. As the network scales toward 1,000+ validator nodes, the volume of model weight updates will grow significantly — reliable decentralised storage becomes increasingly critical at scale.
Success looks like. A live InfiniteZero DevNet where all model weight updates are stored and retrieved via Filecoin, with no centralised pinning dependency. Open source implementation available for other projects to build on.
Outcomes
Primary deliverable. Full integration of Filecoin as the persistent storage layer for InfiniteZero model weight updates — replacing centralised IPFS pinning with Filecoin storage deals, implemented and documented in the open source DevNet repository.
Specific functionality delivered:
- Filecoin storage deal creation for encrypted model weight updates at each training round
- Automated retrieval of model weights by aggregator nodes via Filecoin/IPFS
- CIDv1 compatibility throughout the pipeline
- Documentation and integration guide for other projects seeking to implement similar architecture
Success metrics:
- 100% of model weight updates in the DevNet stored via Filecoin storage deals
- Zero loss of model updates between training rounds
- Integration documented and open source
- At least one additional project implementing the reference architecture within 12 months
Data Onboarding
Model weight updates are small — typically kilobytes to low megabytes per update depending on model size. Projected onboarding based on current DevNet scale and planned growth:
- Month 1: ~1GB (current DevNet scale, ~50 validators)
- Month 3: ~10GB (expanded validator set, multiple training rounds per day)
- Month 6: ~50GB (partner ecosystem pilots begin, multiple models in training)
- Month 12: ~200GB (1,000+ validator nodes, multiple active models)
Adoption, Reach, and Growth Strategies
Target audience. Developers and validators building on the InfiniteZero Network — currently spanning Japan, the USA, Canada, and Europe. Secondary audience: developers in the federated learning and privacy-preserving ML space seeking decentralised storage solutions for model weight persistence.
Current community. Active DevNet with up to 50 validator nodes running at any one time. Backed by Edge City SHIFT Grants (Vitalik Buterin), Cosmos Institute, and the Foresight Institute Berlin node. Recognised by UC Berkeley RDI and the Decentralised Research Center (Ethereum Foundation). $165k raised in non-dilutive funding.
Onboarding strategy. The integration will be documented as a reference implementation in the InfiniteZero open source repository. Speaking engagement at Foresight Institute Berlin (June 2026) will present the implementation to a longtermist and decentralised technology audience. Partner ecosystem onboarding — beginning with distributed solar networks and fitness wearables — will bring additional data contributors onto the network throughout 2026.
Development Roadmap
Milestone 1 — Filecoin Storage Integration (Month 1–2)
Replace centralised IPFS pinning with Filecoin storage deals for model weight updates. Implement automated storage deal creation in the dincli tooling. Test with live DevNet validators.
- Team: Abraham Nash (coordination), Umer Majeed (protocol engineering)
- Deliverables: Filecoin storage deal integration in dincli, tested on live DevNet
- Funding: $8,000
- Completion: Month 2
Milestone 2 — CIDv1 Migration and Retrieval Optimisation (Month 2–3)
Migrate all IPFS uploads to CIDv1 for full Filecoin compatibility. Optimise retrieval latency for aggregator nodes. Document integration architecture.
- Team: Umer Majeed (engineering), Abraham Nash (documentation)
- Deliverables: CIDv1 migration complete, retrieval optimised, integration guide published
- Funding: $5,000
- Completion: Month 3
Milestone 3 — Partner Ecosystem Integration and Reference Architecture (Month 3–6)
Extend Filecoin integration to cover partner ecosystem pilot data flows. Publish reference architecture documentation for other projects.
- Team: Full team
- Deliverables: Partner ecosystem pilots running on Filecoin storage, reference architecture published open source
- Funding: $5,000
- Completion: Month 6
Total Budget Requested
| Milestone # |
Description |
Deliverables |
Completion Date |
Funding |
| 1 |
Filecoin storage deal integration |
dincli integration, live DevNet testing |
Month 2 |
$8,000 |
| 2 |
CIDv1 migration and retrieval optimisation |
CIDv1 complete, integration guide |
Month 3 |
$5,000 |
| 3 |
Partner ecosystem integration and reference architecture |
Partner pilots live, reference docs published |
Month 6 |
$5,000 |
| Total |
|
|
|
$18,000 |
Maintenance and Upgrade Plans
The Filecoin integration will be maintained as a core component of the InfiniteZero protocol. As the network scales toward 1,000+ validators and multiple active training models, storage requirements will grow and the integration will be optimised accordingly. All code is open source under MIT/Apache-2 dual-license and maintained in the InfiniteZero Foundation GitHub repositories. Long-term sustainability is built into the protocol — network fees route to the community DAO, funding ongoing maintenance.
Team
Team Members
- Abraham Nash
- Umer Majeed
- Tudor Cebere
Team Member LinkedIn Profiles
Team Website
https://github.com/InfiniteZeroFoundation
Relevant Experience
Abraham Nash — PhD candidate, University of Oxford, Division of Human-Centered Computing, under Sir Nigel Shadbolt and Emeritus Faculty Sir Tim Berners-Lee. Founded InfiniteZero during doctoral research. Protocol Labs residency alumnus.
Umer Majeed — PhD in Computer Science and Engineering, Kyung Hee University, South Korea. Specialist in federated learning, privacy-preserving computation, and blockchain technologies. Protocol engineer for InfiniteZero.
Tudor Cebere — PhD researcher, Inria France. Associate at Harvard's John A. Paulson School of Engineering and Applied Sciences (OpenDP project). Former core OpenMined contributor. Lead author, PETS 2026 — differential privacy auditing methodology.
Team code repositories
Additional Information
Learned about the Open Grants Program through the Filecoin ecosystem and Protocol Labs network connections established during Abraham Nash's Protocol Labs residency, and through ongoing conversations with Filecoin ProPGF stewards.
Contact: abrahamnash@protonmail.com
InfiniteZero is one of the few live, open source, trustless federated learning protocols running on Ethereum. The Filecoin integration represents a natural and valuable extension of the protocol's commitment to fully decentralised infrastructure — and a meaningful demonstration of Filecoin's utility in the emerging decentralised AI sector.
Project Name: InfiniteZero Network — Filecoin/IPFS Integration for Federated AI Training Infrastructure
Proposal Category: Integrations
Individual or Entity Name: Individual — Abraham Nash (on behalf of the InfiniteZero Foundation team)
Proposer: InfiniteZeroFoundation
Project Repo(s)
Filecoin ecosystem affiliations: Abraham Nash completed a Protocol Labs residency. No current active work relationship with Protocol Labs, Filecoin Foundation, or affiliated organisations.
Technical Sponsor: Pending — in conversation with Filecoin ProPGF stewards.
Do you agree to open source all work under MIT/Apache-2 dual-license?: Yes.
Project Summary
InfiniteZero is a trustless, Ethereum-based protocol for a global AI commons — a decentralised network where communities collectively train AI models while raw data never leaves the device. Participants run lightweight validator nodes that help train, aggregate, and validate model updates. Only anonymised, encrypted model weight updates are shared across the network.
Data is the lifeblood of AI. Without it moving freely and reliably through the system, the training pipeline closes down. The InfiniteZero protocol currently uses IPFS for model weight storage and distribution between training rounds — validator nodes push encrypted model updates to IPFS, where they are retrieved by aggregators for the next round. Reliable, persistent, and decentralised storage of these model weights is critical to protocol integrity. Without it, model updates can be lost between rounds, breaking the training pipeline and invalidating contributor work.
This proposal seeks funding to build out the Filecoin integration as a core infrastructure layer for the InfiniteZero protocol — replacing the current centralised IPFS pinning solution with a fully decentralised, trustless, and cryptographically verified alternative. This work demonstrates a novel, real-world application of Filecoin in live federated learning infrastructure, opening a new category of Filecoin usage in the rapidly growing decentralised AI sector.
Impact
Pain points addressed. The InfiniteZero protocol generates encrypted model weight updates at each training round. These updates must persist reliably between rounds — loss of a model update breaks the training pipeline and invalidates that round's contributions. Current IPFS pinning via centralised services introduces a dependency that is inconsistent with the protocol's trustless architecture. Filecoin's cryptographic storage proofs (Proof of Replication and Proof of Spacetime) provide the guarantees the protocol requires.
Benefits of getting this right. A successful integration demonstrates that Filecoin is a viable storage layer for live AI training infrastructure — not just static data archiving. This opens a new category of Filecoin usage in the federated learning and decentralised AI sector. The open source implementation serves as a reference architecture for other projects seeking to build privacy-preserving, decentralised AI training systems on Filecoin/IPFS.
Risks of not getting this right. Without persistent, trustless storage, the InfiniteZero protocol retains a centralised dependency that undermines its core value proposition. As the network scales toward 1,000+ validator nodes, the volume of model weight updates will grow significantly — reliable decentralised storage becomes increasingly critical at scale.
Success looks like. A live InfiniteZero DevNet where all model weight updates are stored and retrieved via Filecoin, with no centralised pinning dependency. Open source implementation available for other projects to build on.
Outcomes
Primary deliverable. Full integration of Filecoin as the persistent storage layer for InfiniteZero model weight updates — replacing centralised IPFS pinning with Filecoin storage deals, implemented and documented in the open source DevNet repository.
Specific functionality delivered:
Success metrics:
Data Onboarding
Model weight updates are small — typically kilobytes to low megabytes per update depending on model size. Projected onboarding based on current DevNet scale and planned growth:
Adoption, Reach, and Growth Strategies
Target audience. Developers and validators building on the InfiniteZero Network — currently spanning Japan, the USA, Canada, and Europe. Secondary audience: developers in the federated learning and privacy-preserving ML space seeking decentralised storage solutions for model weight persistence.
Current community. Active DevNet with up to 50 validator nodes running at any one time. Backed by Edge City SHIFT Grants (Vitalik Buterin), Cosmos Institute, and the Foresight Institute Berlin node. Recognised by UC Berkeley RDI and the Decentralised Research Center (Ethereum Foundation). $165k raised in non-dilutive funding.
Onboarding strategy. The integration will be documented as a reference implementation in the InfiniteZero open source repository. Speaking engagement at Foresight Institute Berlin (June 2026) will present the implementation to a longtermist and decentralised technology audience. Partner ecosystem onboarding — beginning with distributed solar networks and fitness wearables — will bring additional data contributors onto the network throughout 2026.
Development Roadmap
Milestone 1 — Filecoin Storage Integration (Month 1–2)
Replace centralised IPFS pinning with Filecoin storage deals for model weight updates. Implement automated storage deal creation in the dincli tooling. Test with live DevNet validators.
Milestone 2 — CIDv1 Migration and Retrieval Optimisation (Month 2–3)
Migrate all IPFS uploads to CIDv1 for full Filecoin compatibility. Optimise retrieval latency for aggregator nodes. Document integration architecture.
Milestone 3 — Partner Ecosystem Integration and Reference Architecture (Month 3–6)
Extend Filecoin integration to cover partner ecosystem pilot data flows. Publish reference architecture documentation for other projects.
Total Budget Requested
Maintenance and Upgrade Plans
The Filecoin integration will be maintained as a core component of the InfiniteZero protocol. As the network scales toward 1,000+ validators and multiple active training models, storage requirements will grow and the integration will be optimised accordingly. All code is open source under MIT/Apache-2 dual-license and maintained in the InfiniteZero Foundation GitHub repositories. Long-term sustainability is built into the protocol — network fees route to the community DAO, funding ongoing maintenance.
Team
Team Members
Team Member LinkedIn Profiles
Team Website
https://github.com/InfiniteZeroFoundation
Relevant Experience
Abraham Nash — PhD candidate, University of Oxford, Division of Human-Centered Computing, under Sir Nigel Shadbolt and Emeritus Faculty Sir Tim Berners-Lee. Founded InfiniteZero during doctoral research. Protocol Labs residency alumnus.
Umer Majeed — PhD in Computer Science and Engineering, Kyung Hee University, South Korea. Specialist in federated learning, privacy-preserving computation, and blockchain technologies. Protocol engineer for InfiniteZero.
Tudor Cebere — PhD researcher, Inria France. Associate at Harvard's John A. Paulson School of Engineering and Applied Sciences (OpenDP project). Former core OpenMined contributor. Lead author, PETS 2026 — differential privacy auditing methodology.
Team code repositories
Additional Information
Learned about the Open Grants Program through the Filecoin ecosystem and Protocol Labs network connections established during Abraham Nash's Protocol Labs residency, and through ongoing conversations with Filecoin ProPGF stewards.
Contact: abrahamnash@protonmail.com
InfiniteZero is one of the few live, open source, trustless federated learning protocols running on Ethereum. The Filecoin integration represents a natural and valuable extension of the protocol's commitment to fully decentralised infrastructure — and a meaningful demonstration of Filecoin's utility in the emerging decentralised AI sector.