P112/01
Mumbai → Nashik · MH
NH48 monsoon flood
Causal GNN reads monsoon idx 0.81 + JNPT util 91% → ranks NH60 reroute (12% residual risk).
73% disruption probability · 24h before
SmartChain System®
Five-layer causal pipeline
EST. 2025
GG_Hustlers · GSC
Resilient Logistics · Dynamic Optimisation
Next-gen supply
chains.
The industrialised, causal AI system that brings certainty to logistics. Detect disruptions before they happen. Reroute in milliseconds. Speak to the operator in Hindi.
P.001
Hero
scroll · 06 sections
Causal disruption detection◆Multilingual alerting · Hindi◆Decentralised MARL◆Neural combinatorial routing◆Built on Google◆SDG 8 · 9 · 11◆Causal disruption detection◆Multilingual alerting · Hindi◆Decentralised MARL◆Neural combinatorial routing◆Built on Google◆SDG 8 · 9 · 11◆Causal disruption detection◆Multilingual alerting · Hindi◆Decentralised MARL◆Neural combinatorial routing◆Built on Google◆SDG 8 · 9 · 11◆
[01]Approach
Industrialised vision
and digital control
to execute with certainty.
Governable, replicable, and sustainable processes — assembled from four state-of-the-art models into one continuous pipeline running on Google infrastructure.
/01Causal GNN
Causal, not correlational
A learned DAG over weather, port and traffic variables. Outputs probability + root cause + ranked interventions — not just a score.
Learn more→/02Dec-POMDP · QMIX
Decentralised by design
Each truck, warehouse and carrier is an agent. Centralised critic, decentralised execution. Linear scaling, no bottleneck.
Learn more→/03Gemini · Firebase
Built for the last mile
WhatsApp-first. Hindi · Tamil · Marathi. Confidence-gated handoff to a human. One-tap approval. No dashboard required.
Learn more→$1.5T global annual loss · 24h late before alert · 1M+ concurrent shipments
P.002 / Approach
Causal GNN◆Temporal Fusion Transformer◆Dec-POMDP + QMIX◆POMO + GNN encoder◆Gemini · Firebase · Vertex AI◆Causal GNN◆Temporal Fusion Transformer◆Dec-POMDP + QMIX◆POMO + GNN encoder◆Gemini · Firebase · Vertex AI◆Causal GNN◆Temporal Fusion Transformer◆Dec-POMDP + QMIX◆POMO + GNN encoder◆Gemini · Firebase · Vertex AI◆
[02]System
Five layers.
One pipeline.
Four decision problems, four mathematical structures. Each layer feeds the next — TFT priorities bias POMO attention, Causal GNN risks become edge penalties, QMIX consumes the belief state.
/01
Layer 01 · Disruption Detection
Causal GNN
Probability + root cause + ranked interventions over a learned DAG.
/02
Layer 02 · Demand Forecasting
Temporal Fusion Transformer
7 · 14 · 30-day node forecasts with attention over weather, prices, holidays, news sentiment.
/03
Layer 03 · Resource Allocation
Dec-POMDP + QMIX
Multi-agent reinforcement learning. Centralised training, decentralised execution.
/04
Layer 04 · Route Optimisation
POMO + GNN encoder
Neural combinatorial optimisation over stochastic edge weights from Google Maps.
/05
NH48 जल्द जाम। NH60 से रूट?
✓ Approved · POMO reroute
Layer 05 · Decision & Alerting
Gemini · Human-in-loop
Confidence < 0.75 → Hindi · Tamil · Marathi summary via Firebase push. One-tap approval.
Cambridge AI Lab 2025 · Kwon et al. POMO · CDec-POMDP · TFT
P.003 / System
[03]Cases · Real disruptions
Operating across
India's volatile corridors.
P100/02
Navi Mumbai · MH
JNPT port congestion
QMIX reassigns 40% of carriers from JNPT to Mundra; POMO regenerates 217 routes in 850 ms.
−34% average delay
P158/03
Chennai → Bengaluru · TN/KA
Chennai cyclone corridor
Gemini drafts a Tamil summary with three options. Operator approves backup carrier in one tap.
Tamil alert · 47s response
P148/04
Nashik · MH
Cold-chain spoilage save
Hindi WhatsApp alert at 11:43 PM. Priya approves NH60 reroute. Onions arrive on time.
₹40,000 per shipment · −41% loss
P204/05
Delhi → Jaipur · DL/RJ
Delhi fog corridor
Pre-positioned inventory at Jaipur node based on attention weights from prior fog patterns.
TFT 30-day forecast
/end500+ shipments
500+ shipments
managed in real time.
Time-lapse simulation of Maharashtra road network. Three disruptions caught the human operator would have missed.
← scroll horizontally · pinned
P.004 / Cases
[04]The hero persona
Priya runs cold-chain
in Nashik.
Three active shipments on NH48. At 11:42 PM our system reads monsoon pressure 0.81 and JNPT utilisation 91%. At 11:43 PM she gets a Hindi WhatsApp. One tap reroutes via NH60. Saves ₹40,000 in spoilage.
47s
operator response (was 3h)
−34%
average delay
−41%
spoilage loss
HindiTamilMarathiWhatsAppOne-tap approve
Hindi · Tamil · Marathi|SDG 8 · 9 · 11
11:43
SC
SmartChain
online · Gemini agent
⚠ Disruption alert
आपका शिपमेंट NH48 पर 18–24 घंटे देरी से पहुँच सकता है।
73% probability · monsoon + JNPT 91%
सुझाव: NH60 से रूट बदलें। नया जोखिम 12%। ETA +1 घंटा।
Approved
Reroute confirmed
POMO computed new path · QMIX reassigned carrier #C-204 · ETA 14:20
The emotional hook · before a single algorithm is mentioned
P.005 / Persona
[05]Build with us
Build with an
industrialised &
causal system.
Open the live dashboard. Watch the pipeline run on real shipments, real probabilities, real Hindi alerts — backed by Cloud Run, not a mockup.
/01
Gemini
Multilingual NLG · Hindi · Tamil · Marathi
/02
Maps
Distance Matrix → POMO edges
/03
Firebase
Realtime DB · push to phones
/04
Vertex AI
TFT · Causal GNN · POMO endpoints
/05
Cloud Run
Pipeline orchestration & gate
08
Decent Work & Economic Growth
09
Industry, Innovation & Infrastructure
11
Sustainable Cities & Communities