Auralink is an edge-deployed, agentic software system for EV-charging infrastructure — a safety-relevant, standards-governed domain. This page presents it as a worked reference architecture for Physical AI at production scale on infrastructure the operator owns. It is the founder's own venture, used here as rigorous architecture and safety proof — not a productivity or speed flex, and not an external client engagement.
Last reviewed: June 2026
Auralink is an edge-deployed, agentic software system for EV-charging infrastructure, built by the founder of Hyperion Consulting. Its charger OS is aligned to ISO 15118-20 and OCPP. At approximately 1.7 million lines of first-party code, 400+ microservices, and ~20 AI agents, it serves as a reference architecture for Physical AI at production scale on sovereign infrastructure in a safety-relevant domain. On this page it is presented as architecture and safety evidence — engineering evidence, not certification.
Most discussion of agentic AI lives in demos and benchmarks. Auralink is different: it is a real, operating system in EV-charging infrastructure — a domain that is physical, safety-relevant, and governed by published standards. That makes it a useful reference. The patterns it embodies are not aspirational; they have been carried through at the scale of a production codebase.
The thesis of this page is narrow and deliberate. Auralink is presented as a reference architecture — a worked example of how to architect Physical AI for a standards-governed physical domain — and as safety and architecture proof. It is not presented as a speed story, a velocity metric, or a customer case study. The figures on this page (code scale, microservice count, agent count) are measures of architectural depth, not of how fast anything was built.
Two properties make the reference credible. First, sovereignty: Auralink runs on infrastructure the operator owns, with no mandatory external cloud dependency in the control path. Second, edge-first operation: it is designed to keep working when connectivity is degraded or absent. In a physical domain, both are safety-relevant architectural choices, not deployment conveniences.
One honesty note, stated up front and again below: Auralink is the founder's own venture. It is used here as a rigorous architecture and safety reference. It is not an external client engagement, and no external client names or results are claimed anywhere on this page.
Not a demo or benchmark — a production codebase governing physical charging infrastructure.
EV charging is safety-relevant and governed by ISO 15118-20 and OCPP.
Runs on infrastructure the operator owns; no mandatory external cloud in the control path.
Designed to keep operating under degraded or absent connectivity.
Auralink's architecture is a fan-out of small, independently deployable units running at the edge, with an agentic layer that reasons over them inside a bounded envelope. The numbers below describe architectural surface area and depth — they are scale and design facts, not performance or speed claims.
On the agentic layer: the ~20 agents assist operation — they sense, reason, and arbitrate actions within bounded authority. The engineering team designs, reviews, and owns the system. The agents do not unilaterally build or ship it; they act inside an envelope the architecture defines.
Auralink is a first-party codebase of approximately 1.7 million lines, built AI-native. This figure is presented as a measure of architectural surface area and system depth — not as a productivity or speed claim. A system of this size in a safety-relevant domain demands disciplined module boundaries, explicit interfaces, and an enforced separation between autonomy and enforcement.
The system is decomposed into more than 400 microservices — measured as 477 Dockerfiles across 491 Go modules. This is a fan-out architecture: small, independently deployable units with narrow responsibilities, communicating over well-defined contracts. The microservice boundary is what makes the system observable, individually testable, and recoverable at the unit level.
Approximately 20 AI agents operate within Auralink. The agents assist operation — sensing, reasoning over telemetry, proposing and arbitrating actions within bounded authority. They do not unilaterally build or ship the system; the engineering team designs, reviews, and owns it. The agents act inside an envelope that the architecture, not the agent, defines.
Auralink is edge-deployed: it runs at or near the charging infrastructure it governs, and it is designed to keep operating when external connectivity is degraded or absent. Edge-first is not a deployment convenience — in a standards-governed physical domain it is a safety and continuity property. The system does not depend on a round-trip to a distant cloud to make local decisions.
Hyperion frames Physical AI as six verbs: Sense · Connect · Compute · Reason · Act · Orchestrate. Auralink maps cleanly onto each layer — which is precisely why it works as a reference. The mapping below shows how an edge-first, agentic system for a standards-governed domain populates the stack.
Telemetry from charging hardware, power electronics, session state, and environmental signals is acquired at the edge. In a standards-governed domain, the sensing layer is also where conformance-relevant signals (session negotiation state, fault flags) are captured for downstream reasoning and audit.
Devices and services interoperate over open protocols. Auralink's charger OS is aligned to OCPP (Open Charge Point Protocol) for charge-point communication and to ISO 15118-20 for vehicle-to-infrastructure session negotiation. Connectivity is built around the open standards the domain mandates, not a proprietary bus.
Computation runs on infrastructure the operator owns, at the edge. The ~400-microservice decomposition is the compute substrate: workloads are placed close to the hardware they serve, with no mandatory dependency on an external cloud control plane to execute local logic.
The ~20-agent layer reasons over sensed state — diagnosing conditions, proposing actions, and arbitrating between competing objectives within bounded authority. Reasoning is constrained by explicit policy; the agents operate inside an envelope defined by the system architecture and the relevant standards, not outside it.
Actions reach physical charging infrastructure through controlled interfaces. As with any safety-relevant physical system, the actuation path is where bounding and enforcement matter most: action authority is scoped, and protocol-level constraints (ISO 15118-20 / OCPP) frame what the system is permitted to command.
Across many edge sites, microservices, and agents, an orchestration layer coordinates lifecycle, deployment, versioning, and recovery. Orchestration is what turns hundreds of independent units into a coherent, operable system that can be reasoned about, updated, and rolled back as a whole.
If you are designing an edge-deployed, agentic system for a safety-relevant domain — and need the architecture to stand up to a safety argument — the patterns on this page are the starting point. Hyperion works alongside your team on Physical AI deployment and on the engineering evidence a safety case requires.
EV charging is a standards-governed domain. Auralink's charger OS is aligned to ISO 15118-20 and OCPP, and the agentic layer operates inside a bounded envelope defined by the architecture and those standards. Building on open, documented contracts is what makes a system in this domain auditable — and what makes a later safety argument tractable.
Important boundary: a notified body assigns any safety rating. The material on this page is engineering evidence describing how the system is architected for a standards-governed domain — it is not a certification, and it does not assert one. For the engagement that assembles a safety argument, see Hyperion's Safety-Case & Certification.
Road vehicles — Vehicle to grid communication interface — Part 20
ISO 15118-20 specifies the second-generation application and presentation layer for communication between an electric vehicle and the charging infrastructure, including session negotiation, secure communication, and bidirectional power transfer scenarios. It is the standard that governs how a charger and a vehicle establish and run a charging session.
In Auralink
Auralink's charger OS is aligned to ISO 15118-20 at the session layer. For a reference architecture, the relevant point is structural: the system is designed so that vehicle-to-infrastructure negotiation conforms to the governing standard rather than to an ad-hoc protocol.
Open Charge Point Protocol
OCPP is the open, vendor-neutral protocol for communication between charge points and a central management system. It is the de facto interoperability standard for EV-charging networks, covering transactions, configuration, firmware management, and diagnostics across heterogeneous hardware.
In Auralink
Auralink's charger OS is aligned to OCPP. Building on OCPP keeps the architecture interoperable and auditable: the management-plane behaviour follows a documented open contract, which is a precondition for any later conformance or safety-case work.
The agentic layer reasons and proposes within bounded authority. The boundaries — what the system is permitted to command, under which conditions — are defined by the architecture and the governing standards, not by the agents themselves. This separation is the architectural pattern that makes an agentic system defensible in a safety-relevant domain.
Aligning the charger OS to ISO 15118-20 and OCPP means the externally observable behaviour follows documented, auditable contracts. A safety case is far easier to construct on top of standard-conforming interfaces than on top of proprietary, undocumented ones.
Because Auralink runs at the edge and is offline-capable, local decisions do not depend on a distant cloud round-trip. In a physical domain, the ability to keep operating safely under degraded connectivity is itself a safety-relevant architectural property.
Auralink runs on infrastructure the operator owns. There is no mandatory external cloud dependency in the control path. Sovereignty here is not a marketing posture — it is an architecture choice that keeps the operator in control of the system that governs their physical assets.
The evidence below is presented honestly and in context. The headline figure comes from a preprint — not a refereed paper — and the scale figures are architecture and depth proof, not speed claims.
78% incident resolution preprint 2603.08736, "Autonomous Edge-Deployed AI Agents for EV Charging Infrastructure" (February 2026). This is a preprint — neither refereed nor formally published.
The founder's arXiv preprint 2603.08736, "Autonomous Edge-Deployed AI Agents for EV Charging Infrastructure" (February 2026), reports a 78% incident resolution figure for the agentic edge system. This is a preprint — not a refereed or formally published paper. It is cited here as architectural evidence: a documented account of how an edge-deployed agentic system behaves in a safety-relevant domain.
The ~1.7M-line first-party codebase is presented as evidence of architectural depth and production scale, not as a speed or velocity claim. It demonstrates that the patterns described on this page — edge-first decomposition, an agentic layer inside an enforced envelope, standard-aligned interfaces — have been carried through at the scale of a real, operating system rather than a prototype.
EV-charging infrastructure is a safety-relevant, standards-governed domain. The relevant evidence is not a single benchmark but the fact that the architecture was built to operate inside that domain's constraints — ISO 15118-20 and OCPP alignment, edge-first operation, and bounded agent authority — rather than retrofitted to them.
Auralink is the founder's own venture. On this page it is used as a rigorous architecture and safety reference — a worked example of Physical AI at production scale in a standards-governed, safety-relevant domain.
It is not an external client engagement, and no external client names, logos, or results are claimed anywhere on this page. Every figure here — code scale, microservice count, agent count, and the 78% incident-resolution figure — describes Auralink itself, sourced as stated.
The 78% figure is from a preprint, not a refereed publication. And to restate the standards boundary: a notified body assigns any safety rating — the material here is engineering evidence, not certification.
No. Auralink is the founder's own venture. It is presented on this page as a rigorous architecture and safety reference — a worked example of Physical AI at production scale on sovereign infrastructure in a safety-relevant domain. It is not an external client engagement, and no external client names or results are claimed here.
Auralink operates in EV-charging infrastructure, a standards-governed domain. Its charger OS is aligned to ISO 15118-20 (vehicle-to-infrastructure communication, including session negotiation) and to OCPP (Open Charge Point Protocol, the open charge-point management protocol). These open standards are the interoperability and conformance substrate the architecture is built on.
The relevant Hyperion engagements are Physical AI Deployment (edge-first architecture, agentic layers, and integration for physical systems) and Safety-Case & Certification support (assembling the engineering evidence a safety argument requires). The right starting point is a conversation about your domain, its governing standards, and where an edge-deployed system fits. See /services/physical-ai-deployment and /services/safety-case-certification.
No — it is not. The 78% incident-resolution figure comes from the founder's arXiv preprint 2603.08736, "Autonomous Edge-Deployed AI Agents for EV Charging Infrastructure" (February 2026). It is a preprint, not a refereed or formally published paper. The preprint is linked in the Sources section so the claim can be read in its original context.
Cherifi, M. (2026). "Autonomous Edge-Deployed AI Agents for EV Charging Infrastructure." https://arxiv.org/abs/2603.08736
Context: arXiv preprint 2603.08736 (February 2026). Founder's preprint — not refereed or formally published — reporting a 78% incident-resolution figure for an edge-deployed agentic system in EV-charging infrastructure. Cited on this page as architectural evidence.
ISO (2022). "ISO 15118-20: Road vehicles — Vehicle to grid communication interface — Part 20."
Context: International standard specifying the second-generation application and presentation layer for EV-to-infrastructure communication, including session negotiation, secure communication, and bidirectional power transfer.
Open Charge Alliance (2024). "OCPP — Open Charge Point Protocol."
Context: Open, vendor-neutral protocol for communication between EV charge points and a central management system. The de facto interoperability standard for EV-charging networks.
If you are building Physical AI for a safety-relevant, standards-governed domain — and you want the architecture to stand up to scrutiny — start with a conversation. Hyperion brings production experience in edge-deployed, agentic systems and works alongside your team on deployment and on the engineering evidence a safety case requires.
Founder & AI Strategy Lead
Mohammed Cherifi is the founder of Hyperion Consulting and the founder of Auralink, an edge-deployed agentic system for EV-charging infrastructure. With 17+ years in automotive and embedded systems engineering, he specialises in Physical AI deployment for safety-relevant, standards-governed domains.
Edge-first architecture, agentic layers, and integration for safety-relevant physical systems
Assembling the engineering evidence a safety argument requires — a notified body assigns any rating
Closing the sim-to-real gap: physics simulation through on-robot edge inference
The Physical AI Stack — Sense, Connect, Compute, Reason, Act, Orchestrate