Skip to content
ACAnkit Chatriki
Contact
Back to my workFILE 02 OF 02
SOLO BUILD

Sutra

Demand forecasting & supply chain analytics copilot

An analytics copilot for independent restaurants that turns POS, labor, inventory and demand data into clear reorder and prep decisions.

WEEKLY OVERSTOCK CAUGHT77%Hot Chicken Slider Duo, before it expired
UNITS FLAGGED BEFORE EXPIRY154With options to move or resize the order

  1. The problem

    Independent restaurants run on thin margins, but sales, labor, food cost and inventory live in separate systems, so over-ordering and waste go unnoticed until month-end.

  2. The solution

    A FastAPI and PostgreSQL copilot that runs a Meta Prophet forecasting service on daily POS sales, explodes forecast demand through recipe BOMs (28 menu items to 65 SKUs), and sets reorders and daily prep with EOQ, safety stock, reorder points and a newsvendor model.

  3. Real example

    Demand for the Hot Chicken Slider Duo dropped 36% in 30 days, but the standing order still delivered 288 buns a week against 67 needed. Sutra flagged the 77% surplus, with 154 units four days from expiry, and recommended a promotion and a smaller standing order.

Watch the demo3:07 · PRODUCT WALKTHROUGH, NO AUDIO
Forecast vs actuals, BOM network28 ITEMS TO 65 SKUS
Prophet demand forecast with a prediction band against actual daily sales, showing weekly seasonality. Tick marks along the base show the 40 rolling-origin backtest folds, and the shaded zone marks the forecast horizon.40 rolling-origin foldsHorizonBill-of-materials network: 28 menu items on the left, each connected to the SKUs it uses among 65 SKUs on the right.28 menu items65 SKUs
Prophet demand forecast with a prediction band against actual daily sales, showing weekly seasonality. Tick marks along the base show the 40 rolling-origin backtest folds, and the shaded zone marks the forecast horizon.40 rolling-origin foldsHorizonBill-of-materials network: 28 menu items on the left, each connected to the SKUs it uses among 65 SKUs on the right.28 menu items65 SKUs

Prophet demand forecast with a prediction band against actual daily sales, showing weekly seasonality. Tick marks along the base show the 40 rolling-origin backtest folds, and the shaded zone marks the forecast horizon. Bill-of-materials network: 28 menu items on the left, each connected to the SKUs it uses among 65 SKUs on the right.

01The problem

What if a restaurant owner could ask a question about their business and get one clear answer connecting sales, costs, labor, inventory and demand?

POSLaborOpExRecipe costsInventoryDemand
One clear answerSutra

02The principle

Owners understand their business better than any model can. Sutra surfaces the exposure and lays out the options.

03How it works

1 ProfitabilityEstimated continuously
2 ForecastProphet service
3 BOM explosion28 items to 65 SKUs
4 Replenish + prepConstraints and newsvendor
  • Profitability: POS revenue, labor, operating expenses and recipe-level food cost, with no wait for the accounting close.
  • Forecast: trend and day-of-week seasonality from daily sales.
  • Replenishment: safety stock, reorder point, EOQ, shelf life and pack size show which constraint drives a reorder.
  • Daily prep: a newsvendor model balances expected demand, stockout cost and salvage value.

04Caught in the data: the Hot Chicken Slider Duo

Demand −36%Over 30 days
288 buns a weekStanding order, vs 67 needed
77% overstock154 units at risk, 4 days to expiry
AI layerLays out the options
  • Promote or discount the at-risk stock to move it.
  • Resize the standing order so it does not repeat next week.

No single system would have flagged it alone. Sutra put the three together and caught it.

05Under the hood: forecast accuracy

Rolling-origin backtest, mean across folds.

AVERAGE FORECAST MISS13.1%WAPE
21% LESS ERROR THAN A SIMPLE GUESS0.79MASE
BIAS+1.2%Forecast bias
BACKTEST40Rolling-origin folds

MASE under 1.0 beats a naive forecast.

06My role

Solo build. I designed and developed the full system: the forecasting service, the BOM and replenishment pipeline, and the AI insights layer.

Ankit Chatriki
Ankit ChatrikiDEVELOPER & CREATOR

07Stack

PythonFastAPIMeta ProphetSQLPostgreSQLClaude API
© 2026 Ankit ChatrikiBuilt with Next.js, GSAP, LenisBack to top
Back to home