Skip to content

ForecastAI 2026.01 — Mirabelle

ForecastAI Mirabelle is a supply-chain intelligence platform that combines statistical and deep-learning demand forecasting, multi-echelon inventory optimisation (MEIO), supply planning, causal/asset-driven forecasting, automated alerts, and exception management — all within a multi-tenant, pipeline-based architecture.

What Mirabelle Does

Mirabelle ingests your historical demand data and produces:

  • Demand forecasts — 24-week rolling horizon using 10+ statistical and neural models (AutoARIMA, AutoETS, MSTL, Croston, NHITS, NBEATS, PatchTST, TFT, DeepAR, TimesFM).
  • Best-method selection — composite weighted backtesting automatically picks the winning model per SKU.
  • MEIO / Inventory optimisation — safety-stock, reorder points, EOQ exchange curves using fitted demand distributions (Normal, Gamma, Negative Binomial, Lognormal, Weibull).
  • Supply planning — MRP-style supply runs generating BUY / BUILD / REPAIR / TRANSFER orders with pegging, capacity constraints, and inventory projections.
  • Score-based allocation — deterministic priority-based distribution of limited supply across competing demands.
  • Replay engine — iterative re-run of supply + allocation over a configurable historical horizon for fill-rate and stock-evolution analysis.
  • Causal forecasting — asset-usage-driven demand for MRO / aerospace, maintenance scheduling, failure-rate waterfalls.
  • Exceptions & alerts — configurable ontology-based alert rules with workflow (open → reviewed → snoozed → actioned).
  • Multi-tenant SaaS — each customer gets their own PostgreSQL database; a master database holds tenant accounts and the shared JWT secret.

Key Capabilities at a Glance

Capability Where
10 forecasting models files/forecasting/
Rolling-window backtesting files/evaluation/
Composite best-method scoring files/selection/best_method.py
Demand characterisation (intermittency, seasonality, trend) files/characterization/
Distribution fitting for safety stock files/distribution/fitting.py
Multi-echelon inventory optimisation files/meio_runner.py, files/MEIO_v2/
Supply planning engine files/api/supply_router.py
Score-based allocation files/allocation_runner.py
Replay engine files/replay/replay_engine.py
Causal / asset-driven forecasting files/forecasting/causal/
Alert ontology engine files/alerts/engine.py
88+ REST API endpoints files/api/main.py (port 8002)
React 18 SPA files/frontend/ (port 5173)
Multi-tenancy forecastai_master PostgreSQL DB

Full Pipeline Flow

flowchart TD
    ERP["ERP / Source DB\n(demand_db)"]
    ETL["ETL\nfiles/etl/\nWeekly aggregation"]
    CHAR["Characterisation\nfiles/characterization/\nSeasonality · Trend · Intermittency"]
    FORE["Forecasting step\nfiles/forecasting/\n① Outlier detection (IQR · z-score · STL) → demand_corrected\n② StatsForecast · NeuralForecast · TimesFM\n(Dask parallel)"]
    EVAL["Evaluation\nfiles/evaluation/\nRolling-window backtesting"]
    BEST["Best Method Selection\nfiles/selection/best_method.py\nComposite weighted scoring"]
    DIST["Distribution Fitting\nfiles/distribution/fitting.py\nNormal · Gamma · NegBin · Lognormal · Weibull"]
    PG["PostgreSQL\nscenario schema\nTenant database"]
    API["FastAPI\nfiles/api/main.py\nPort 8002 · 200+ endpoints"]
    REACT["React 18 Frontend\nfiles/frontend/\nPort 5173"]
    MEIO["MEIO / Inventory Optimisation\nfiles/meio_runner.py\nSafety stock · EOQ · ROP"]
    SUPPLY["Supply Planning\nfiles/api/supply_router.py\nBUY / BUILD / REPAIR / TRANSFER orders"]
    ALLOC["Allocation Engine\nfiles/allocation_runner.py\nScore-based priority allocation"]
    REPLAY["Replay Engine\nfiles/replay/replay_engine.py\nIterative supply+allocation replay"]
    CAUSAL["Causal Forecasting\nfiles/forecasting/causal/\nAsset-driven · Maintenance"]
    ALERTS["Alert Engine\nfiles/alerts/engine.py\n32 rule ontology"]

    ERP --> ETL --> CHAR --> FORE --> EVAL --> BEST --> DIST --> PG
    PG --> API --> REACT
    DIST --> MEIO --> PG
    PG --> SUPPLY --> PG
    SUPPLY --> ALLOC --> PG
    PG --> REPLAY
    CAUSAL --> PG
    PG --> ALERTS

Documentation Sections

  • User Guide — Overview, architecture, quickstart, daily workflows for demand, supply, and inventory planners
  • Algorithms — Forecasting models, backtesting, distribution fitting, MEIO maths
  • Developer Guide — Backend code, database schema, auth, multi-tenancy, ports
  • Configuration — Parameter sets, segments, pipeline setup
  • Administration — Tenant provisioning, user management, SuperAdmin panel

Quick links