8 AI modules working together as a unified food logistics intelligence system — from demand prediction to surplus redistribution
Built on production-grade open-source stack
Next.js 16 Frontend
SSR + Turbopack, Tailwind CSS, Recharts
PostgreSQL + Prisma
23 models, full relational schema
Python FastAPI ML
Prophet, XGBoost, scikit-learn sidecar
Gemini 2.0 Flash
Menu generation, insight analysis
BPS + BMKG APIs
Live price & weather data sync
QRIS + BI-FAST
Payment settlement & audit trail
Each module feeds into the next — fully automated pipeline
Module 1 of 8
MBG Surplus Redistribution Marketplace
Problem
2,400 tons of MBG leftovers wasted daily — 451K+ tons/year still fit for consumption. No digital platform connects SPPG with 9,100+ panti asuhan and communities.
PADI Solution
AI redistribution platform matching surplus food from 23,000 SPPG to nearby panti asuhan, panti jompo, mosques, and food banks. Claim via QR code, full audit trail per BGN Regulation No. 1/2026.
SPPG reports surplus via PADI-WASTE → auto-create listing in marketplace
AI matches surplus to nearest recipient based on capacity, distance, and food type
Recipient claims portions → generate unique QR code → WhatsApp notification
Pickup verified via QR scan → complete audit trail for BGN compliance
KEY TECHNOLOGIES
0%
Target edible waste from 23,000 SPPG
Module 2 of 8
Demand Forecasting & Price Prediction
Problem
School coordinators currently estimate food needs by guesswork, leading to 20-30% over/under-ordering across 514 kabupaten.
PADI Solution
Prophet + TFT hybrid model combines BPS price data, BMKG weather, school enrollment (DAPODIK), and historical consumption to forecast demand per kabupaten with 92% accuracy.
BPS API (prices), BMKG API (weather), DAPODIK (enrollment) → daily sync
Prophet for trend/seasonality + Temporal Fusion Transformer for multi-variate
7/14/30-day demand forecast per kabupaten per commodity with confidence intervals
Gemini 2.0 Flash generates natural-language insights when anomalies detected
KEY TECHNOLOGIES
92%
Forecast accuracy (MAPE < 8%)
Module 4 of 8
Smart Supplier Matching
Problem
Procurement is manual, opaque, and biased toward incumbent distributors. Local UMKM farmers are excluded from the MBG supply chain.
PADI Solution
Multi-factor supplier scoring algorithm (quality, price, distance, reliability, capacity) with geographic optimization — prioritizes local UMKM and farmer cooperatives within 50km radius.
PADI-MENU generates ingredient list → quantities calculated from enrollment × recipe
Search supplier registry within configurable radius, filter by product, capacity, certifications
Weighted scoring: quality (30%), price (25%), distance (20%), reliability (15%), capacity (10%)
Top-N suppliers selected per ingredient, auto-generate purchase orders with QRIS payment links
KEY TECHNOLOGIES
78%
Orders fulfilled by local UMKM/farmers
Module 5 of 8
Multi-Modal Route Optimization
Problem
Indonesia spans 17,000+ islands. Manual logistics planning results in 30-40% cost overruns and frequent cold-chain breaks, especially for eastern Indonesia routes.
PADI Solution
Constrained optimization across truck, ship, and air modes with cold-chain requirements, port schedules, and real-time weather from BMKG. Minimizes cost × time × spoilage risk.
Build logistics graph: 514 kabupaten as nodes, road/sea/air as weighted edges
Cold-chain requirements, port schedules, ferry capacity, perishability windows
Modified Dijkstra with multi-objective: cost (weight 0.4), time (0.3), spoilage risk (0.3)
GPS + temperature IoT sensors → real-time re-routing if cold-chain breach detected
KEY TECHNOLOGIES
-32%
Avg cost reduction vs manual routing
Module 6 of 8
Waste Prediction & Redistribution
Problem
Indonesia wastes 23-48 million tons of food annually. School-level waste data is untracked — surplus food goes to landfill instead of nearby panti asuhan or food banks.
PADI Solution
ML model predicts waste risk per school based on menu, weather, day-of-week, and historical patterns. Surplus auto-matched to nearest redistribution points (panti asuhan, food banks, PMI).
Daily meal reports: prepared vs consumed, per school, per menu item
Random Forest classifies waste risk: LOW (<10%), MEDIUM (10-20%), HIGH (>20%)
Portion adjustment suggestions sent to kitchen staff via WhatsApp/dashboard
Surplus matched to nearest recipient, tracked donation for tax/compliance
KEY TECHNOLOGIES
34%
Waste redistributed instead of discarded
Module 7 of 8
QRIS Payment & Supply Chain Finance
Problem
Government food procurement payments take 30-90 days via traditional banking, forcing UMKM suppliers to take loans. Payment trail is opaque, enabling corruption.
PADI Solution
QRIS-based direct payment to farmer/UMKM wallets with BI-auditable trail. UMKM credit scoring unlocks micro-financing. Digital Rupiah concept for programmable budget enforcement.
Purchase order generated by PADI-MATCH → QRIS payment link created
Coordinator scans QRIS → funds settle to supplier in <5 seconds via BI-FAST
Transaction history builds UMKM credit score (300-850) across 5 factors
Credit score enables micro-loans from partnering banks for capacity expansion
KEY TECHNOLOGIES
<5s
Payment settlement time (vs 30-90 days)
Module 8 of 8
National Monitoring Dashboard
Problem
No unified view of MBG program performance across 514 kabupaten. Decision-makers rely on monthly Excel reports — too slow for early intervention.
PADI Solution
Real-time national dashboard with choropleth map, KPI cards, alert system, and drill-down to kabupaten/school level. Powered by aggregated data from all 7 other modules.
All module outputs feed into unified metrics store (meals, waste, spend, alerts)
514-kabupaten choropleth map, time-series charts, KPI sparklines
Rules engine: price spikes >20%, waste >15%, cold-chain >4°C, stock <3-day threshold
Auto-generated weekly PDF reports for BGN, KEMENKO PMK, and Bank Indonesia
KEY TECHNOLOGIES
Real-time
Data freshness (vs monthly Excel)
See all 8 modules in action with real dummy data — or walk through the end-to-end demo scenario