The narrative behind the numbers — recent signals clustered into storylines, each with its evidence one click away. Composed by the LLM from live mentions; regenerated every few hours.
Generated by the LLM from live alerts + internal context
104 weeks of brand demand learned against real weather, live sentiment and the procurement book. Every arrow below is an elasticity the model recovered itself — not a hardcoded assumption. Stress it, forecast with it, and make it prove itself on past shocks.
Read left to right: history trains the model → sources feed the drivers → drivers move weekly demand → the supply chain fulfils it → P&L. Click any node for its links; every arrow declares whether it is learned, book, assumed or user-added.
Click any node to inspect its links. learned from data · procurement book · assumed · accounting · signal feed · trained on
Shock the drivers; learned elasticities + the procurement book propagate to rupees. Bands widen where the model fits less well (effect × (1−R²)).
Weekly demand, actual vs model-fitted, around the two real shocks in the spine. If the dashed line tracks the solid one, the elasticities can be trusted forward.
Baseline = same week last year; effects from learned elasticities. Week 1 temperature is the real Open-Meteo forecast. Dotted red = plant capacity.
Open-Meteo, live & keyless · 22 state anchors · next-7d mean Tmax, badge = anomaly vs the last 30 days. Connected: states aggregate to the regional demand reads below (anomaly × learned elasticity), drive week 1 of the forward ribbon, and raise a ticker alert past ±1.5°.
Highest-impact competitor signals from live news, LLM-scored
Category = competitor, newest first
Computed from ingested coverage volume per brand group
Each product fights its own war — click one to open its battleground: target group, markets and the competitor set from Parle Agro's product matrix; sentiment and share of voice computed live from the last 30 days.
Compose the scenario from movers and levers — or type anything. Answered by the LLM using internal sales + learned elasticities; every number arrives with its arithmetic, auto-audited.
Point the 9-language mention store at any launch — yours or a rival's. Buzz, sentiment and murmurs go live from the store the moment you add keywords (include native scripts for vernacular lead time). Outlet and trial telemetry stays with HANA/Salesforce and is labeled seeded until those connect.
Illustrative scenario — seeded, not live data. A hypothetical launch showing how signals would arrive day by day. The grey line under each murmur names the feed it would come from; see README for which of those are actually connected.
Live coverage and sentiment by state, fused with sales performance. Click a state to pin it.
Revenue (HANA-shaped) × distribution (Salesforce-shaped) × share × sentiment
Radar seeded with verified research (Aug 2026); live high-impact regulatory items are added automatically and tagged LIVE. Below it, the individual scored mentions the storylines are composed from.
Live, ranked by LLM impact score
Live, ranked by LLM impact score
Care lines + social + marketplace reviews (demo data — point at Salesforce cases)
The working detail behind the twin's cost nodes: live signals per driver, a procurement read composed by the LLM, and the mechanical sensitivity from the seeded P&L (directional until HANA connects). Not part of the main navigation — reach it from the Twin's header.
STATE
Mention share within this battleground only
Newest first — our brand and every rival in the set
MODEL ASSUMPTIONS · THE EDITABLE BOOK
Everything the model assumes rather than learns — yours to set until HANA/WMS connect. Capacity shares renormalize to sum to 1. Changes apply to the graph, the Stress Lab, the outlook and the drift watchdog immediately.
GRAPH NODES · ADD YOUR OWN
Extend any column — a source you track, a driver, a channel, a depot. User nodes are annotations on the graph until their connector lands; they do not enter the stress arithmetic and are tagged CUSTOM.