SequenxAI
ZHContact

SequenxAI Platform · Architecture deep-dive

From field datato quantile forecasts

This page covers how the SequenxAI Platform is built: where time series come in, how they are cleaned, which interface calls the models, and how results flow back into your systems. What the platform does, and its boundaries, are on the platform page.

SequenxAI Platform architectureTime series from data sources enter the cleaning pipeline, then the unified forecasting API calls the general time-series model (domain-tuned models are planned) and returns quantile forecasts and a data_quality report to the customer's dispatch, risk and reporting systems. Cleaning, API, models and outputs can run on onshore nodes or in a private deployment.Onshore nodes / private deploymentSOURCESCLEANINGAPIMODELSOUTPUTSSYSTEMSCSV / databaseSensors / IoTBusiness systemsCleaning pipelineAlign · de-duplicateFill short gapsFlag outliers onlyForecasting APIAuth · quotasIdempotency · auditGeneral modelReady on cold startDomain modelsPlanned, by industryQuantile outputsP10 · P50 · P90data_quality reportDispatchRiskReporting / BI

Call path

Request → clean → quality report → forecast & intervals

  1. 01

    Long-table input

    Submit structured long-table series (covariate columns are accepted; model use is planned); timestamps need not be perfectly regular, since cleaning snaps small offsets to the grid.

  2. 02

    Cleaning pipeline

    Timestamp alignment, de-duplication, short-gap interpolation, and frequency inference for messy field data; outliers are flagged, never changed.

  3. 03

    data_quality

    Responses carry auditable quality reports; failures return structured error codes for ops and compliance.

  4. 04

    Forecast & intervals

    Point forecasts plus quantile intervals (e.g. P10–P90).

Components

Four building blocks

  • 01

    One gateway

    Every call enters through the same gateway: authentication, quotas, rate limits and idempotency are settled before any compute, and usage is metered per request.

  • 02

    Control and compute apart

    Accounts, keys and metering run on the control plane; cleaning and inference run on separate compute nodes, so each scales on its own.

  • 03

    Data residency

    Sensitive data stays onshore on domestic nodes; on-premises deployment can be agreed per project.

  • 04

    Traceable failures

    Every response carries a data_quality report; failures return structured error codes, and call logs are visible in the platform.

Planned

Planned,not yet delivered

Planned

  • Smart routing: after cleaning, choose the general or an industry model per scenario and calibrate intervals
  • Same-scenario public comparisons
  • Domain-tuned models join the same API after same-scenario benchmarks

Enterprise evaluation

Evaluate this architecturein your environment

When you need onshore nodes or a private deployment, agree the evaluation scope and delivery with the team.

Apply for evaluationSee platform capabilities