CodeGang

ARM Tech ERP

Live in production

IN · Kolkata · Trade & Logistics

CementBook + TruckBook + FinanceBook — a complete 3-module ERP suite for a cement, fleet & cold-materials trader. Five deep-learning modules on top: LSTM demand forecasting, autoencoder anomaly detection, trip-profitability scoring, invoice OCR, and dealer-risk classification.

Visit the live system ↗Build something like this

CEMENTBOOK

Purchases · Sales · Stock · Parties · Payments · Expenses

TRUCKBOOK

Fleet management with per-truck trip logs & profit tracking

FINANCEBOOK

Capital, bank, cash & receivables/payables accounting

LIVE URL

cement-app.vercel.app

Real screens, not mockups

CementBook Dashboard — sales, profit, stock, receivables, payables, capital
CementBook Dashboard — sales, profit, stock, receivables, payables, capital
Purchases — 63 rows, filterable by date, brand & supplier with Excel export
Purchases — 63 rows, filterable by date, brand & supplier with Excel export
TruckBook Dashboard — per-truck P&L, active trucks, total trips, monthly freight
TruckBook Dashboard — per-truck P&L, active trucks, total trips, monthly freight
FinanceBook — Revenue & Capital, bank breakdown, receivables & payables
FinanceBook — Revenue & Capital, bank breakdown, receivables & payables

What the system does

Live profit

Sales − purchases, refreshed on every entry.

Receivables view

Customers who owe us, ranked & filterable.

Payables view

Suppliers we owe, with aging and due dates.

Bank vs Cash

Real-time net cash + bank reconciliation.

Brand mix

Monthly sales breakdown by cement brand & type.

Excel export

One-click export of any module's data.

Five deep-learning modules on one pipeline

LSTM + temporal features

Demand Forecasting

Predicts 7- to 30-day cement-bag demand per brand, godown & region — feeds purchase planning so stockouts and overstocks both drop.

Autoencoder on tx vectors

Anomaly Detection

Flags unusual entries in payments, expenses & bank movements — catches data-entry errors and suspicious activity in near real time.

Tabular DL regressor

Trip Profitability

Scores every potential truck trip on expected net profit using freight, fuel, route & driver features — guides dispatch in TruckBook.

CNN backbone + Transformer

Invoice OCR

Reads supplier bills from a phone camera and auto-fills purchase entries — cuts manual import time per bill from minutes to seconds.

Tabular DL classifier

Dealer Risk Scoring

Scores each dealer/party on dormancy and default risk, surfacing accounts that need a follow-up before they go cold.

How it is put together

Sources

Supabase (Postgres) — Sales & purchases ledgers, bank & cash transactions, truck trip logs, supplier invoices (images)

Feature Pipeline

Nightly Airflow batch jobs — time-series feature builders, tabular cleaning & encoding, image preprocessing (OCR), feature store in Postgres

Model Layer

LSTM — demand forecasting · Autoencoder — anomalies · Tabular DL — trip profit · CNN+Transformer — OCR · Tabular DL — dealer risk

Serving

FastAPI inference service · Cached predictions in Redis · Surfaced in CementBook UI · Alerts via dashboard widgets · Excel-export ready outputs

Stack

  • Next.js
  • React
  • Supabase
  • PostgreSQL
  • FastAPI
  • LSTM
  • TensorFlow
  • Airflow
  • Redis
  • Vercel

Where it stands

Modules
CementBook · TruckBook · FinanceBook
AI modules
5 deep-learning modules on one pipeline
Stack
Next.js · Supabase · SSO Auth · Excel export · Vercel deploy

Need a system like ARM Tech ERP?

Tell us what you are trying to ship. The engineer who built this one answers.

Start a projectsupport@codegang.online