Skip to main content

Enterprise architecture resources

Review the existing solution architecture outlines alongside the product and operational guides they reference.

Solution architecture outlines

These diagrams and descriptions are drawn from the existing solution pages. Open a solution for its full scope and linked product references.

AI API

Use every published model with OpenAI-compatible Chat Completions or Claude Messages. Find exact identifiers, client examples, authentication and protocol-specific guidance.

Open product guide

Request and response path

Inference request
  1. Customer-controlled Application client

    The integrating application forms and sends the request.

    submits a request
    Platform component Authentication and entitlement

    Request identity and access are evaluated at the service boundary.

  2. Platform component Authentication and entitlement

    Request identity and access are evaluated at the service boundary.

    authorized request proceeds
    Platform component Model routing

    The request is directed according to the selected model and available integration.

  3. Platform component Model routing

    The request is directed according to the selected model and available integration.

    routes to selected provider
    External system Upstream provider

    A third-party provider handles the selected model request.

  4. External system Upstream provider

    A third-party provider handles the selected model request.

    returns provider response
    Customer-controlled Application response

    The client receives the response for its own application handling.

Conceptual request path. Authentication, entitlement, routing, and upstream behavior depend on the selected integration and provider.

Architecture references

Model Training API

Follow the complete TrainingProject workflow, explore the backward-compatible Python client and API, and build a reproducible local training run.

Open product guide

FusionPointAI Training SDK session and runtime flow

Authenticate and prepare authorized data
  1. Customer-controlled Customer API key

    The TrainingClient sends the customer API key as a bearer credential.

    authenticates the TrainingClient API request
    Platform component Authenticated paid job session

    The platform checks job ownership, training-SDK product, purchased runtime, session expiry, and current paid reference access.

  2. Platform component Authenticated paid job session

    The platform checks job ownership, training-SDK product, purchased runtime, session expiry, and current paid reference access.

    checks current paid reference-model access
    Platform component Selected paid reference model

    The platform uses the configured, currently authorized reference model to generate examples; it does not supply student-model weights.

  3. Platform component Selected paid reference model

    The platform uses the configured, currently authorized reference model to generate examples; it does not supply student-model weights.

    generates bounded training and new held-out validation examples
    Platform component Platform data preparation

    The platform generates bounded training examples and distinct held-out validation examples from the configured objective and criteria.

  4. Platform component Authenticated paid job session

    The platform checks job ownership, training-SDK product, purchased runtime, session expiry, and current paid reference access.

    authorizes the owned, configured job session
    Platform component Platform data preparation

    The platform generates bounded training examples and distinct held-out validation examples from the configured objective and criteria.

  5. Platform component Platform data preparation

    The platform generates bounded training examples and distinct held-out validation examples from the configured objective and criteria.

    validates and stores the train and validation batches
    Platform component Train and validation datasets

    Session endpoints serve train and validation JSONL with manifests. This session flow does not transfer a final-test split.

Load the student and execute training
  1. Platform component Train and validation datasets

    Session endpoints serve train and validation JSONL with manifests. This session flow does not transfer a final-test split.

    SDK downloads authenticated splits and verifies manifest size, rows, and SHA-256
    FusionPointAI software FusionPointAI Training SDK

    FusionPointAI provides the TrainingClient and trainer software; the SDK runs inside the selected runtime.

  2. FusionPointAI software FusionPointAI Training SDK

    FusionPointAI provides the TrainingClient and trainer software; the SDK runs inside the selected runtime.

    loads the selected student checkpoint
    Customer-controlled Selected student checkpoint

    The SDK loads the configured student repository and revision for training.

  3. FusionPointAI software FusionPointAI Training SDK

    FusionPointAI provides the TrainingClient and trainer software; the SDK runs inside the selected runtime.

    FusionPointAI SDK executes optimization in the configured runtime
    Selected runtime operator Training runtime and operator

    The configured target is FusionPointAI Hosting, a customer datacenter, or compatible Colab; its operator supplies compute.

Report evaluation and retain the runtime artifact
  1. Selected runtime operator Training runtime and operator

    The configured target is FusionPointAI Hosting, a customer datacenter, or compatible Colab; its operator supplies compute.

    posts runtime-reported training and evaluation events
    Platform component Progress and completion report

    The platform records SDK-reported events and validates submitted evaluation fields; verification is runtime-reported, not independent.

  2. Selected runtime operator Training runtime and operator

    The configured target is FusionPointAI Hosting, a customer datacenter, or compatible Colab; its operator supplies compute.

    saves the actual model artifact in the runtime
    Selected runtime operator Runtime-local model artifact

    The SDK saves the artifact in the runtime. The platform stores its reported name and SHA-256 metadata, not the model bytes.

  3. Selected runtime operator Runtime-local model artifact

    The SDK saves the artifact in the runtime. The platform stores its reported name and SHA-256 metadata, not the model bytes.

    returns artifact name and hash metadata, not model bytes
    Platform component Progress and completion report

    The platform records SDK-reported events and validates submitted evaluation fields; verification is runtime-reported, not independent.

The authenticated TrainingClient opens a paid job session for its configured runtime. FusionPointAI prepares and serves train and held-out validation splits; the FusionPointAI Training SDK runs in FusionPointAI Hosting, a customer datacenter, or compatible Colab runtime. Completion records runtime-reported evaluation and artifact metadata; model bytes remain in the runtime. A separate final-test split is not transferred by this session API.

GPU Hosting

Evaluate GPU infrastructure using a concrete model, memory profile, runtime, concurrency, and delivery target. Browse current hardware and hosting catalogues instead of relying on generic capacity or performance assumptions.

Open product guide

Hosting request to workload operations

Scope and deploy
  1. Customer-controlled Hosting request

    The customer describes workload, deployment, and capacity needs.

    informs scope review
    Defined by agreed scope Commercial allocation

    Scope and capacity assumptions are reviewed before an allocation is agreed.

  2. Defined by agreed scope Commercial allocation

    Scope and capacity assumptions are reviewed before an allocation is agreed.

    defines the agreed allocation
    Defined by agreed scope Workload deployment

    The workload is placed in the environment defined by the agreed deployment model.

Operate and deliver
  1. Defined by agreed scope Workload deployment

    The workload is placed in the environment defined by the agreed deployment model.

    is observed within scope
    Defined by agreed scope Monitoring

    Operational visibility follows the responsibilities in the selected scope.

  2. Defined by agreed scope Monitoring

    Operational visibility follows the responsibilities in the selected scope.

    supports operational handoff
    Defined by agreed scope Delivery and handoff

    Access and operational handoff follow the agreed delivery process.

Illustrative service path. Capacity, placement, monitoring, and delivery are subject to product availability and the agreed deployment scope.

Open solution details: Choose compute from workload and verified availability

Architecture references

Logical model
Deployment
Integration
Security

Model Engineering

Assess model transformation options against release evidence, target hardware, runtime compatibility, language needs, and delivery mode. Work from a specific source model and acceptance criteria rather than assuming every transformation is available.

Open product guide

Artifact engineering and validation path

Engineering engagement
  1. Customer-controlled Input artifact

    The customer identifies the model or artifact and its permitted use.

    is assessed for scope
    Defined by agreed scope Scope and quote

    The requested work, dependencies, and commercial scope are documented.

  2. Defined by agreed scope Scope and quote

    The requested work, dependencies, and commercial scope are documented.

    proceeds after scope agreement
    Platform component Model engineering

    Work is performed within the confirmed technical scope.

  3. Platform component Model engineering

    Work is performed within the confirmed technical scope.

    is checked against criteria
    Defined by agreed scope Validation

    Results are reviewed against agreed acceptance criteria.

  4. Defined by agreed scope Validation

    Results are reviewed against agreed acceptance criteria.

    is handed off when accepted
    Platform component Delivery

    The agreed outputs and handoff information are provided.

A scoped engineering sequence; quote, validation criteria, deliverables, and responsibilities are confirmed for each engagement.

Open solution details: Adapt model releases to a documented production target

Architecture references

Enterprise Banking

Evaluate a banking blockchain environment for institutions that need explicit asset policies, transaction controls, governance responsibilities, and a planned path from a non-production evaluation to an agreed operating model.

Open product guide

Banking transaction, fee, and responsibility map

Banking application and asset controls
  1. Customer-controlled Bank applications

    Customer channels and internal systems initiate authorized integration requests.

    integrates through the selected API
    Defined by agreed scope Banking API integration

    Selected APIs connect bank systems to the network functions in scope.

  2. Defined by agreed scope Banking API integration

    Selected APIs connect bank systems to the network functions in scope.

    submits actions for configured authorization
    Defined by agreed scope Governance and authorization

    Configured authorities and policies govern permitted asset operations.

  3. Defined by agreed scope Governance and authorization

    Configured authorities and policies govern permitted asset operations.

    applies configured authority to asset actions
    Defined by agreed scope Native asset controls

    The configured asset lifecycle can include mint, burn, and restriction actions.

  4. Defined by agreed scope Native asset controls

    The configured asset lifecycle can include mint, burn, and restriction actions.

    submits mint, burn, or restriction changes
    Defined by agreed scope Network validators

    Validators process network transactions under the selected network configuration.

  5. Defined by agreed scope Network validators

    Validators process network transactions under the selected network configuration.

    makes recorded network data available to the explorer
    Defined by agreed scope Explorer and reporting

    Network data is presented through the configured explorer and integrations.

Technical gas and customer fees
  1. Customer-controlled Bank applications

    Customer channels and internal systems initiate authorized integration requests.

    may use configured technical gas sponsorship
    Defined by agreed scope Paymaster

    Technical gas sponsorship is a transaction mechanism, not the bank’s customer-fee policy.

  2. Defined by agreed scope Paymaster

    Technical gas sponsorship is a transaction mechanism, not the bank’s customer-fee policy.

    handles technical transaction gas
    Defined by agreed scope Network validators

    Validators process network transactions under the selected network configuration.

  3. Customer-controlled Bank fee policy

    The bank determines customer-facing charges and disclosures separately.

    the bank defines customer-facing fee policy separately
    Customer-controlled Bank customer fees

    Customer fees are governed by the bank’s own products, policy, and disclosures.

Keys and deployment responsibility
  1. Defined by agreed scope Managed or bank-hosted deployment

    The selected deployment model defines hosting and operational boundaries.

    sets the agreed key-control boundary
    Defined by agreed scope Key responsibility

    Key custody and administration remain with the party identified in the agreed scope.

  2. Defined by agreed scope Managed or bank-hosted deployment

    The selected deployment model defines hosting and operational boundaries.

    sets the agreed hosting and operations boundary
    Defined by agreed scope Hosting and operations

    Operational ownership follows the managed or bank-hosted scope.

Conceptual integration map. Technical gas handling is distinct from the bank’s customer-fee policy. Hosting, key control, and operational responsibilities depend on the agreed deployment scope.

Open solution details: Modernize financial workflows with institution-owned controls

Architecture references

Managed EVM Networks

Assess managed network, infrastructure, and operational product options while keeping protocol choices, validator duties, access, upgrades, incident handling, and customer responsibilities explicit.

Open product guide

Application access and network service relationships

Application path
  1. Customer-controlled Applications

    Customer applications use the configured endpoint and network.

    sends application requests
    Platform component RPC access

    Application requests enter through the configured RPC interface.

  2. Platform component RPC access

    Application requests enter through the configured RPC interface.

    connects to configured network services
    Defined by agreed scope Validators

    Validator components participate in the selected EVM network.

Network operations
  1. Defined by agreed scope Validators

    Validator components participate in the selected EVM network.

    operates under network governance
    Defined by agreed scope Governance

    Network governance follows the selected network and its defined authorities.

  2. Defined by agreed scope Validators

    Validator components participate in the selected EVM network.

    provides network data for exploration
    Platform component Explorer

    An explorer presents network information for review.

  3. Defined by agreed scope Validators

    Validator components participate in the selected EVM network.

    has an operational scope
    Defined by agreed scope Operations

    Operational responsibilities are established in the agreed service scope.

Conceptual component map for a managed EVM service. It names no live network and does not imply a particular availability or performance level.

Open solution details: Operate network services with defined responsibility boundaries

Architecture references

Related product and operating guides

The links above lead to existing documentation chapters and product references; each remains the source for its technical details.