convez.ai
Plugin
bkumars22

SupplyChainPlatformProject

bkumars22/SupplyChainPlatformProject

Ver no GitHub
InicialPluginScore · de 0 a 100

Code

Instalação

$npx convez-ai bkumars22/SupplyChainPlatformProject

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
57%
Último commit
há 1 mês
Versão
—
Issues abertas
0
Contribuidores
2

README

A tradução para o português deste README ainda está na fila.

Supply Chain Intelligence Platform (SCIP)

CI Tests LLM Eval License Java Python React


The supply chain platform that replaces your spreadsheets — built by one engineer in 4 weeks using AI.

Live Demo: https://scip.railway.app (Coming Soon) Login: kumar / Kumar@2026


The Business Problem

Most mid-market manufacturers run their supply chains in Excel. A full-scale implementation in top industries typically costs between $200,000 and $2,000,000 and requires approximately 18 months to complete. The gap in between — where most companies actually live — is where SCIP operates.

If your team is doing any of these things right now, SCIP was built for you:

Pain PointHow Teams Handle It TodayWhat SCIP Does Instead
Supplier reliabilityManual scorecards in a shared spreadsheetAutomatic OTD scoring, tier badges, delivery history
Cost change approvalsEmail chains with attachmentsDigital DRAFT → SUBMIT → APPROVE → REJECT workflow
Supply disruption warningsFind out when it's too lateAI flags anomalies before they become problems
Demand forecastingExcel models built by one analyst who might leaveML-powered forecasting with variance tracking
Test and quality visibilityEmail report or separate tool loginLive test dashboard embedded in the product at /tests
Mobile access for managersVPN + laptop requiredFull platform on Android and iOS

Live Demo

Login credentials: kumar / Kumar@2026

Dashboard — Morning KPI view for supply chain managers

Start every day with the numbers that matter: active alerts, pending approvals, at-risk suppliers, and BOM health — all in one screen.

Dashboard


Alerts — AI-detected anomalies before they become problems

IsolationForest ML scans every supplier and cost record continuously. When something looks wrong, Claude AI writes the explanation in plain English — no data science degree required to understand it.

Alerts


Bill of Materials — Multi-level BOM with real-time status

Track every component across every product. Filter by status, drill into detail, and see approval state without calling anyone.

BOM


Cost Records — Digital approval workflow that replaces email chains

Every cost change request moves through a tracked workflow. Nothing gets approved without a record. Nothing gets lost in someone's inbox.

Cost Records


Suppliers — Automatic OTD scoring replaces manual scorecards

Every supplier gets a real on-time delivery score, a tier classification, and a delivery history — updated from the data, not from whoever last edited the spreadsheet.

Suppliers


Forecasts — ML-powered demand forecasting

Demand and cost forecasts with variance analysis. Create draft forecasts, track actuals, and see where the model was wrong — all without a separate BI tool.

Forecasting


AI Engine — Supplier risk scoring at a glance

The AI anomaly dashboard shows which suppliers and cost records triggered risk flags, what score they received, and why. Powered by IsolationForest trained on supply chain data.

AI Engine


User Management — Full RBAC without an IT ticket

Three roles: Administrator, Business Administrator, Read Only. Create users, set passwords, disable accounts, and manage access — all from within the app.

Users


Test Intelligence Dashboard — Live quality metrics embedded in the product

Go to /tests in the app. See all 41 test results, 14 API health checks, and every known bug — without logging into any external tool. This is what quality visibility looks like when it's built into the product instead of bolted on.

Test Dashboard

Playwright Results

API Health

Known Bugs

Live Results


Who Uses This

Supply Chain Manager Arrives Monday morning, opens the dashboard, sees 3 active alerts and 2 suppliers in the amber tier. Clicks into the alert, reads the AI explanation, dismisses the one that's a known quarterly pattern, escalates the other. Does all of this before the first meeting of the day — no spreadsheets opened.

Finance Analyst Gets notified of a cost record pending approval. Opens it, sees the full audit trail from DRAFT through SUBMIT, compares against the forecast variance, approves it. The whole workflow is tracked. No email chain. No version confusion.

Procurement Manager Walking the factory floor before a supplier meeting. Pulls up the supplier scorecard on the mobile app: OTD 87%, Tier 2, last 10 deliveries. Has the conversation with real numbers, not memory.

QA Lead Wants to know if the last deployment broke anything. Goes to /tests in the app. Sees 36/41 tests passing, 14 API health checks green, 5 known bugs with severity and fix instructions. No Jira login. No Jenkins. No tool-switching.


Architecture

Enterprise-grade stack delivering SAP-level features at startup cost.

┌─────────────────────────────────────────────────────────────┐
│  Clients                                                     │
│  React 18 Web (port 3000) · React Native iOS/Android        │
│  Data Portal HTML · Test Intelligence Dashboard             │
└──────────────────────┬──────────────────────────────────────┘
                       │ JWT Bearer Token · REST
┌──────────────────────▼──────────────────────────────────────┐
│  Spring Boot API  (port 8089/supchain)                      │
│  80+ endpoints · JWT auth · BCrypt · RBAC · XSS filter      │
│  JPA/Hibernate · H2 (dev) · PostgreSQL (prod)               │
└────────┬─────────────────────────────┬───────────────────────┘
         │ JPA                          │ HTTP
┌────────▼────────┐         ┌──────────▼──────────────────────┐
│  Database        │         │  Python AI Service (port 8001)   │
│  H2 / PostgreSQL │         │  FastAPI · IsolationForest ML    │
│  80+ tables      │         │  DistilBERT failure prediction   │
│  Flyway migrations│        │  Claude AI alert explanations    │
└──────────────────┘         │  LangGraph agent workflows       │
                             └─────────────────────────────────┘

The Java backend handles all business logic, auth, and data persistence. The Python AI service is a separate microservice that the backend calls over HTTP — keeping the ML stack decoupled from the JVM. The React web app and React Native mobile app share the same JWT-authenticated REST API.


AI Capabilities

IsolationForest Supplier Risk Scoring

An unsupervised ML model (scikit-learn) that scores every supplier based on their on-time delivery rate, cost variance, and order history. Suppliers that behave differently from the baseline cluster get flagged. No labeled training data required — the model learns what "normal" looks like and alerts on deviations.

DistilBERT Failure Prediction

A fine-tuned transformer model trained on 949 real CI failure logs. Given a test run output, it predicts the failure category before a human reads the stack trace. Reduces triage time from minutes to seconds.

Claude AI Natural Language Alerts

When IsolationForest flags an anomaly, SCIP calls Claude AI to generate a one-paragraph explanation in plain English — written for a supply chain manager, not a data scientist. "Supplier XYZ's last 3 deliveries averaged 4 days late, with cost variance 18% above their 90-day baseline" instead of a raw anomaly score.

LLM Evaluation Harness

20 structured test cases measuring the quality of every AI prompt response on a 0–1 scale. Current average: 0.918/1.0. Tests cover factual accuracy, format compliance, hallucination rate, and business relevance. All 20 pass. This is production-grade LLM quality control, not vibe checks.

LangGraph Agent Workflows

The autonomous build-fix-test pipeline uses LangGraph to orchestrate multi-step agent workflows: diagnose build failure → apply fix → rebuild → run tests → commit → report. The agent makes decisions at each node based on the output of the previous step.


Tech Stack

LayerTechnology
BackendJava 17 · Spring Boot 3 · Maven · JWT (HMAC-SHA256) · BCrypt · JPA/Hibernate
FrontendReact 18 · React Router · Axios · CSS-in-JS
MobileReact Native · Expo SDK · Shared JWT auth
AI/MLPython FastAPI · IsolationForest · scikit-learn · DistilBERT · Claude AI API · LangGraph
TestingPlaywright TypeScript · Custom persistent reporter · Test Intelligence Dashboard
DevOpsDocker Compose · GitHub Actions CI/CD · Railway (production)
DatabaseH2 (dev, in-memory) · PostgreSQL (prod) · Flyway migrations · Full seed via data.sql

Quick Start

Prerequisites

  • Java 17+
  • Node.js 18+
  • Python 3.9+
  • Maven 3.9+
  • Docker Desktop (optional)

Option 1 — Run locally (3 terminals)

# 1. Backend
git clone https://github.com/bkumars22/SupplyChainPlatformProject.git
cd SupplyChainPlatformProject
mvn package -Dmaven.test.skip=true
java -jar target/pcm-0.0.1-SNAPSHOT.war

# 2. Web app (new terminal)
cd scweb && npm install && npm start

# 3. Python AI service (new terminal)
cd ai-service && pip install -r requirements.txt
uvicorn main:app --port 8001

Option 2 — Docker Compose (single command)

docker compose up -d

Option 3 — Smart Fix Agent (Windows, recommended for first run)

Set-ExecutionPolicy Bypass -Scope Process
.\agents\master_fix.ps1

The agent detects build failures, applies fixes, starts all services, and verifies all APIs automatically.

Service URLs

ServiceURL
Web Apphttp://localhost:3000
Test Dashboardhttp://localhost:3000/tests
Spring Boot APIhttp://localhost:8089/supchain
Swagger UIhttp://localhost:8089/supchain/swagger-ui/index.html
H2 Consolehttp://localhost:8089/supchain/h2-console
Python AI Servicehttp://localhost:8001

API Explorer (Swagger UI)

Swagger UI

All 80+ endpoints documented and testable from the browser. Click Authorize, paste a JWT token from POST /api/auth/login, and test any endpoint live.

API GroupWhat It Does
AuthenticationJWT login, current user
DashboardAggregated KPI summary
AlertsActive alerts, unread count, dismiss
Bill of MaterialsList, detail, stats, filter by status
Cost RecordsPaginated list, stats, filter by status
ForecastsList, detail, stats, create DRAFT
ItemsSearch, create, AVL entries
SuppliersSupplier list with OTD scores
User ManagementCRUD users, list roles
AI EngineRisk scoring, anomaly detection
Password ManagementChange password, admin reset

Test Results

Playwright End-to-End Tests — 41 cases, 36 passing

ModuleTestsResult
Authentication77 pass ✅
Item Master66 pass ✅
Bill of Materials77 pass ✅
Cost Records65 pass · 1 known bug (P1)
Suppliers32 pass · 1 known bug (P3)
User Management65 pass · 1 known bug (P2)
Alerts44 pass ✅
Dashboard22 pass ✅

LLM Evaluation — 20 cases, 20 passing, avg score 0.918/1.0

20 structured test cases measuring prompt quality across factual accuracy, format compliance, hallucination rate, and business relevance. All pass.

API Health Checks — 14 live checks

All core API endpoints verified on every run: status code, response time, payload shape.

Bugs Found Through Black Box Testing

IDSeverityDescriptionStatus
BB-AUTH-02P0 CriticalAny password accepted — BCrypt check skipped when password field was null✅ Fixed — seed users now have BCrypt-hashed passwords
BB-COST-02P1 HighPOST /api/cost-records returns 405 — @PostMapping missing from controllerOpen — use /api/costs (full CRUD available)
BB-ITEM-04P2 MediumPOST /api/items returns 500 — NPE in service layer on items with no AVL data✅ Fixed — null guard added in ItemManagementService
BB-USR-03P2 MediumPOST /api/admin/users returns 500 — roleKey FK resolution failsOpen
BB-SUP-03P3 LowSupplier tier badge doesn't match OTD score in seed dataOpen

Finding and documenting a P0 auth bypass through black box testing — before any real users touched the system — is the point of this discipline.


Key Metrics

┌─────────────────────────────────────────────────────────────────┐
│                                                                  │
│   13+ years     ·    4 weeks to build    ·   5 engineering      │
│   experience         with AI assistance      domains            │
│                                                                  │
│   80+ API       ·    41 automated        ·   0.918 / 1.0        │
│   endpoints          tests                   LLM eval score      │
│                                                                  │
│   819 Java      ·    25+ autonomous      ·   10 Fortune 500      │
│   source files       agents                  clients served      │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Business Value Statement

A typical enterprise ERP implementation costs between $200,000 and $2,000,000 and takes around 18 months to complete. This platform delivers the core supply chain workflows in a single command.

It was built to demonstrate that one AI-augmented engineer — with the right depth of domain knowledge and engineering discipline — can deliver what previously required a team of 20 and a budget with six zeros.

The platform is not a demo. It has a real auth system, a real approval workflow, real ML models, real test coverage, and real bugs found and fixed. That is what production-grade looks like.


Project Structure

SupplyChainPlatformProject/
│
├── src/                          # Java Spring Boot backend
│   └── main/java/
│       ├── com/test/pcm/         # Core supply chain domain
│       └── com/scplatform/api/   # REST controllers, JWT, security
│
├── scweb/                        # React 18 web application
│   └── src/
│       ├── pages/                # 9 page components
│       └── api.js                # All API call functions
│
├── SupplyChainApp/               # React Native mobile app (iOS + Android)
│
├── ai-service/                   # Python FastAPI AI microservice
│   ├── main.py                   # IsolationForest, DistilBERT, Claude AI
│   └── eval/                     # LLM evaluation harness (20 test cases)
│
├── playwright-tests/             # Playwright TypeScript test suite
│   ├── tests/scip.spec.ts        # 41 test cases
│   └── persistent-reporter.js    # Writes results to React public/
│
├── docs/
│   ├── ARCHITECTURE.md
│   ├── API_REFERENCE.md
│   ├── BUSINESS_REQUIREMENTS.md
│   ├── USE_CASES.md
│   └── scip_data_portal.html     # Standalone data management portal
│
├── agents/                       # Autonomous fix and deploy agents
│   ├── master_fix.ps1
│   └── scip_orchestrator.ps1
│
├── .github/workflows/
│   ├── ci.yml                    # Run tests on every push
│   └── cd.yml                    # Deploy to Railway on merge
│
└── docker-compose.yml            # 5 services: DB, Redis, API, AI, Web

Documentation

DocumentWhat's In It
Architecture GuideFull system architecture, auth flow, data flows, port reference
API ReferenceAll 80+ endpoints with request/response examples
Business RequirementsBRD with 20 use cases, UC-01 through UC-20
Data PortalStandalone browser tool — create data, run workflows, test APIs

Built By

Kumara Swamy — Staff SDET · AI Development Lead · Supply Chain Domain Expert

13 years building enterprise supply chain software for Fortune 500 manufacturers. This project demonstrates full-stack delivery capability beyond the QA domain — from database schema through ML models through mobile app through CI pipeline.


License

MIT License — see LICENSE for details.

Copyright © 2026 Kumara Swamy — github.com/bkumars22

Ver o original no GitHub

Bom para

Relacionados