Ai B2B Solutions | AI Final Year Project Report

Complete Technical Paper Content — Ai B2B Solutions

1. Executive Abstract, Problem Formulation & Background

Autonomous AI-Driven B2B Deep Research & Hyper-Personalized Outreach Engine An enterprise-grade autonomous intelligence pipeline built in n8n that monitors inbound B2B prospects, orchestrates multi-source web scraping and news research, runs structured LLM reasoning (DeepSeek via OpenRouter), enriches Google Sheets CRM, and crafts contextualized sales outreach with human-in-the-loop safeguards. 🎯 Executive Overview Traditional B2B prospecting suffers from two extremes: generic cold email templates that ruin sender reputation, or hours of manual research per account that throttles sales velocity. This pipeline acts as an Autonomous Senior B2B Sales Engineer : Detects : Ingests new prospective leads (Company & Domain) in real-time via Google Sheets triggers. Researches : Concurrently queries Tavily AI for recent corporate press releases and scrapes the prospect's live website to extract headers and core offerings. Reasons : Feeds ground-truth data into DeepSeek via OpenRouter under strict fact-based system guardrails, performing localized company evaluation and determining the most appropriate executive touchpoint. Enriches & Previews : Writes actionable intelligence back to the CRM sheet and triggers an internal Gmail notification for sales team review prior to final outreach. 🏗️ Architecture & Pipeline Flow flowchart TD A[📊 Google Sheets Trigger\nPolls rowAdded every minute] --> B{🔍 IF Node\nAI_Analysis is empty?} B -- Yes --> C[🔁 Loop Over Items\nSplit in Batches Rate-Limiter] B -- No / Already Processed --> Z[🛑 End / Skip] subgraph Deep_Research ["🌐 Autonomous Deep Research"] C --> D[📡 Tavily Search API\nReal-time Market & Corporate News] D --> E[🕷️ Scrape Target Web\nDirect HTTP Fetch domain] E --> F[🧹 HTML Extractor\nClean semantic tags: p, h1, h2, h3] end subgraph LLM_Reasoning ["🧠 AI Sales Engineering (OpenRouter)"] F --> G[🤖 DeepSeek LLM\nStructured JSON Schema\nFact-Based Guardrails & Dual-Language Routing] end subgraph CRM_and_Outreach ["📤 CRM Synchronization & QA"] G --> H[📝 Google Sheets Update\nSave Analysis, Contact, Subject, HTML Draft] H --> I[✉️ Gmail Notification\nHuman-in-the-Loop Preview & QA] I --> C end Loading ✨ Key Features Multi-Source Real-Time Intelligence : Combines live company website scraping (semantic content extraction) with external news indexing (Tavily AI) to bypass static knowledge cutoffs. Strict Anti-Hallucination Guardrails : The LLM prompt enforces a zero-fabrication policy; outputs are strictly derived from gathered search and scraping payloads. Dynamic Dual-Language Localization : Internal strategic synthesis ( analisis ) is standardized in professional Bahasa Indonesia. Outbound email subjects and cold outreach drafts ( email_subject , email_draft ) automatically adapt to Global English for international entities, or Indonesian for domestic prospects. Schema-Enforced JSON Output : Utilizes JSON mode to guarantee deterministic extraction of kontak , analisis , email_subject , and email_draft . Deduplication & Re-execution Protection : Built-in If condition ensures only rows with empty AI_Analysis are ingested, eliminating infinite execution loops and redundant API costs. Human-in-the-Loop (HITL) Quality Control : Rather than blindly sending cold emails to prospects, the system routes styled email drafts to internal sales review, safeguarding brand credibility. 🛠️ Tech Stack & Integrations Layer Technology Purpose Orchestration n8n (v1.x / Latest) Workflow automation and state control Search Intelligence Tavily AI Search Real-time targeted corporate news and PR retrieval Web Extraction HTTP Request + HTML Node Semantic extraction of DOM elements ( p , h1-h3 ) LLM Inference OpenRouter / DeepSeek ( deepseek-v4-flash-0731 ) High-reasoning structured sales synthesis CRM / Database Google Sheets API Real-time trigger and lead enrichment sheet Outbound / Alerting Gmail OAuth2 Review dispatch and team notifications 📋 Google Sheet CRM Schema To utilize this workflow, create a Google Spreadsheet with the following column headers in Row 1: Column Header Description Populated By Company Name of the target company (e.g., Acme Corp ) User / Lead List Domain Target company website domain (e.g., acme.com ) User / Lead List Contact Target executive / decision-maker role AI Generated AI_Analysis Strategic deep analysis of company positioning & synergy AI Generated Email_Subject Personalized cold outreach subject line AI Generated Email_Draft Complete HTML-formatted cold email proposal AI Generated 🚀 Quickstart & Setup Guide 1. Prerequisites Self-hosted n8n (via Docker) or an active n8n Cloud instance. API Key from Tavily . API Key from OpenRouter . Google Cloud Console Project with Google Sheets API and Gmail API enabled (OAuth2). 2. Environment Configuration Clone the repository and set up your local environment file: git clone https://github.com/zaydan28/n8n-ai-driven-b2b-deep-research.git cd n8n-ai-driven-b2b-deep-research cp .env.example .env Fill in your respective keys in .env : TAVILY_API_KEY = tvly-xxxxxxxxxxxxxxxxxxxxxxxxx OPENROUTER_API_KEY = sk-or-v1-xxxxxxxxxxxxxxxxxxxxxxxxx SALES_NOTIFICATION_EMAIL = sales@yourdomain.com 3. Running n8n with Docker (Optional) If self-hosting via Docker Compose: docker run -it --rm \ --name n8n \ -p 5678:5678 \ --env-file .env \ -v n8n_data:/home/node/.n8n \ docker.n8n.io/n8nio/n8n Access the n8n UI at http://localhost:5678 . 4. Importing the Workflow Open your n8n Dashboard. Click Workflows > Import from File... (or press Ctrl + O ). Select workflow.json from this repository. Update the credentials on the following nodes: Google Sheets Trigger : Connect your Google Sheets OAuth2 credential and select your spreadsheet. Update row in sheet : Connect your Google Sheets OAuth2 credential and link to the same spreadsheet. Send a message : Connect your Gmail OAuth2 credential. Verify that your Tavily and OpenRouter nodes reference your API keys or n8n environment variables. Click Save and toggle the workflow to Active . 🔒 Security & Best Practices Zero Hardcoded Secrets : All private API tokens, spreadsheet IDs, and OAuth references have been sanitized from workflow.json into template placeholders and environment variables. Data Minimization : Website scraping parses only semantic text nodes to optimize token usage and avoid scraping unnecessary personal identifiers. Fail-Safe Rate Limiting : The Loop Over Items (Split in Batches) node processes prospects one-by-one, adhering to external domain and API rate quotas. 👤 Author Zaydan GitHub: @zaydan28 Focus: AI Automation, Agentic Workflows, n8n Systems Engineering 📄 License This project is licensed under the MIT License . Feel free to adapt and use it for your automation architecture.

An enterprise-grade autonomous intelligence pipeline built in n8n that monitors inbound B2B prospects, orchestrates multi-source web scraping and news research, runs structured LLM reasoning (DeepSeek via OpenRouter), enriches Google Sheets CRM, and crafts contextualized sales outreach with human-in-the-loop safeguards.

Traditional B2B prospecting suffers from two extremes: generic cold email templates that ruin sender reputation, or hours of manual research per account that throttles sales velocity.

flowchart TD A[📊 Google Sheets Trigger\nPolls rowAdded every minute] --> B{🔍 IF Node\nAI_Analysis is empty?} B -- Yes --> C[🔁 Loop Over Items\nSplit in Batches Rate-Limiter] B -- No / Already Processed --> Z[🛑 End / Skip] subgraph Deep_Research ["🌐 Autonomous Deep Research"] C --> D[📡 Tavily Search API\nReal-time Market & Corporate News] D --> E[🕷️ Scrape Target Web\nDirect HTTP Fetch domain] E --> F[🧹 HTML Extractor\nClean semantic tags: p, h1, h2, h3] end subgraph LLM_Reasoning ["🧠 AI Sales Engineering (OpenRouter)"] F --> G[🤖 DeepSeek LLM\nStructured JSON Schema\nFact-Based Guardrails & Dual-Language Routing] end subgraph CRM_and_Outreach ["📤 CRM Synchronization & QA"] G --> H[📝 Google Sheets Update\nSave Analysis, Contact, Subject, HTML Draft] H --> I[✉️ Gmail Notification\nHuman-in-the-Loop Preview & QA] I --> C end

To utilize this workflow, create a Google Spreadsheet with the following column headers in Row 1:

2. Related Work, Literature Survey & Algorithmic Foundations

git clone https://github.com/zaydan28/n8n-ai-driven-b2b-deep-research.git cd n8n-ai-driven-b2b-deep-research cp .env.example .env

TAVILY_API_KEY = tvly-xxxxxxxxxxxxxxxxxxxxxxxxx OPENROUTER_API_KEY = sk-or-v1-xxxxxxxxxxxxxxxxxxxxxxxxx SALES_NOTIFICATION_EMAIL = sales@yourdomain.com

docker run -it --rm \ --name n8n \ -p 5678:5678 \ --env-file .env \ -v n8n_data:/home/node/.n8n \ docker.n8n.io/n8nio/n8n

This project is licensed under the MIT License . Feel free to adapt and use it for your automation architecture.

Build B2B Commerce Agents with Synapse AI A comprehensive boilerplate for creating intelligent commerce agents. Get started in minutes with pre-built components, AI integrations, and best practices. 🚀 Quick Links Backend : https://api.mach-x-b2b.app/docs - FastAPI backend with LangGraph orchestration Frontend : https://mach-x-b2b.app - React chat interface Documentation : https://docs.mach-x-b2b.app - Comprehensive documentation site GitHub Repository : View on GitHub 📖 Project Documentation Backend README - Backend setup, architecture, and API documentation Frontend README - Frontend development guide and setup Documentation Site README - Documentation site setup Terraform README - Infrastructure as Code deployment guide Backend Scripts README - Utility scripts and maintenance tools WhatsApp Integration README - WhatsApp Business API integration 🎯 What is SynapseAI? SynapseAI is an intelligent B2B wholesale commerce platform that powers conversational AI-driven ordering, inventory management, and campaign automation for CPG brands. About FlavorFlux (Demo Brand) FlavorFlux is our fictional demo brand - representing an adaptive CPG yogurt drink brand. Throughout this documentation, FlavorFlux serves as the example to showcase SynapseAI's capabilities. Key Features 🤖 Conversational AI Natural language ordering Product search and recommendations Cart management through chat Multi-turn dialogue support 📊 AI-Powered Intelligence Demand forecasting with AWS Forecast Predictive reordering for inventory optimization Campaign template generation with Amazon Bedrock Community sentiment analysis 🛒 B2B Commerce Integration with Commercetools for products, carts, and orders Voucherify for promotions and loyalty programs Real-time inventory tracking Order history and reordering 🎯 Human-in-the-Loop Approval workflows for high-value orders Risk mitigation for unusual patterns Audit trails for compliance Notification systems 🏗️ Technology Stack Backend FastAPI : High-performance Python web framework LangGraph : Stateful AI agent orchestration (SynapseAI core) LiteLLM : Multi-provider LLM interface Redis : Session and cache management MCP : Model Context Protocol for integrations Frontend React 19 : Modern UI library Vite : Lightning-fast build tool TanStack Router : Type-safe routing TanStack Query : Server state management Shadcn/ui : Beautiful component library Tailwind CSS : Utility-first styling Infrastructure AWS ECS Fargate : Serverless containers AWS ALB : Load balancing AWS ElastiCache : Redis cluster AWS ECR : Container registry Terraform : Infrastructure as Code External Services Commercetools : E-commerce platform Voucherify : Promotions engine AWS Bedrock : AI foundation models AWS Forecast : Demand prediction 🚦 Getting Started Ready to dive in? Here's the recommended path: Start with Getting Started - Understand the project and set up your environment Follow Backend Setup - Get the backend running locally Set up the Frontend - Build and run the user interface Review Architecture - Understand how everything fits together Explore User Scenarios - See the platform in action Deploy with Deployment Guide - Take it to production 📊 System Overview flowchart TB Users["👥 Users (Admins, Buyers, Store Reps)"] Frontend["🎨 React Frontend (Chat Interface)"] Backend["⚡ FastAPI Backend (API Gateway)"] Agent["🤖 SynapseAI Agent (LangGraph Orchestration)"] Services["🔧 Services (MCP Clients)"] Commerce["🛒 Composable Core (Commercetools, Voucherify)"] AI["🧠 AWS AI (Bedrock, Forecast)"] Users --> Frontend Frontend --> Backend Backend --> Agent Agent --> Services Services --> Commerce Agent --> AI Loading 🤝 Brought to you by 👥 Know the Team Behind This Tomasz Pindel CEO, Voucherify Subhasri Vadyar Sr Solutions Architect, Valtech Kyle Barz Sr Director Enterprise Architecture, Mars Kerrigan Baron CEO & Founder, Fidget Lab Gireesh Sahukar Sr Ecommerce Technologist Tomasz Kuczmarski Sr Software Developer, Voucherify Tomasz Sikora Software Developer, Voucherify Federico Orlandau Frontend Architect, Valtech Hocine Bedouhene Sr Ecommerce Architect, Valtech Bruno Citroni Full Stack Developer, Mira Commerce Edu De Lisa CTO, Mira Commerce Sergei Ostapenko CEO, Mira Commerce 📝 Documentation Structure Setup Guides Getting Started - Overview and quick start Backend Setup - Backend installation and configuration Frontend Setup - Frontend development guide Deployment - Production deployment with Terraform Architecture & Design Architecture - High-level system architecture C4 Architecture - Detailed C4 model diagrams User Scenarios - User personas and workflows Tools & Utilities Scripts - Maintenance scripts and utilities 🤝 Contributing This is an internal project. For questions or contributions, please contact the development team. 📝 License This project is licensed under the MIT License - see the LICENSE file for details. 🔗 Social Media Twitter : @MachAlliance LinkedIn : MACH Alliance Instagram : @machalliance Platform : SynapseAI v1.0.0 Demo Brand : FlavorFlux Last Updated : November 2025

3. Dataset Ingestion, Preprocessing & Stochastic Augmentations

A comprehensive boilerplate for creating intelligent commerce agents. Get started in minutes with pre-built components, AI integrations, and best practices.

SynapseAI is an intelligent B2B wholesale commerce platform that powers conversational AI-driven ordering, inventory management, and campaign automation for CPG brands.

FlavorFlux is our fictional demo brand - representing an adaptive CPG yogurt drink brand. Throughout this documentation, FlavorFlux serves as the example to showcase SynapseAI's capabilities.

flowchart TB Users["👥 Users (Admins, Buyers, Store Reps)"] Frontend["🎨 React Frontend (Chat Interface)"] Backend["⚡ FastAPI Backend (API Gateway)"] Agent["🤖 SynapseAI Agent (LangGraph Orchestration)"] Services["🔧 Services (MCP Clients)"] Commerce["🛒 Composable Core (Commercetools, Voucherify)"] AI["🧠 AWS AI (Bedrock, Forecast)"] Users --> Frontend Frontend --> Backend Backend --> Agent Agent --> Services Services --> Commerce Agent --> AI

This is an internal project. For questions or contributions, please contact the development team.

4. Neural Network Backbone, Layer Specifications & Loss Formulations

Awesome AI Architect 10x your AI Architect journey with plain‑language, hands‑on guides that blend AI engineering and solution architecture. 📚 Table of Contents 🏗️ Core Topics AI Architecture Patterns - RAG, agents, pipelines, and orchestration Vector Stores & Embeddings - Storage, retrieval, and similarity search Databases for AI - Relational tables, vector indexes, feature stores, and metadata catalogs Serving & Scaling - Inference servers, GPUs, and autoscaling Inference Infrastructure - GPUs, TPUs, quantization, and cost-performance metrics Evaluation & Observability - Testing, monitoring, and cost tracking Guardrails - NeMo Guardrails, Rebuff, and custom policy engines Observability - Prometheus, OpenTelemetry, drift detection, and alerting practices Safety & Security - OWASP LLM Top 10, guardrails, and best practices Red Teaming - Scenario generation, attack simulations, and reporting workflows Security & Compliance - Regulatory requirements, encryption, and model provenance tracking Orchestration Frameworks - LangChain, LangGraph, and workflow tools Code Execution Sandboxes - E2B micro-VMs for isolated tool execution and security Advanced RAG - Multi-hop retrieval, rerankers, and feedback loops Cloud for AI - AWS, Azure, and GCP services and best practices Multi-Cloud - Portability, data residency, failover, and cost arbitrage Model Routing - Dynamically choose the best model for a task Multi-LLM Strategies - Voting, specialists, and cost-aware selection Data Scraping - robots.txt, throttling, and content licensing Data Lakes & Marts - Lakehouse concepts, ETL/ELT pipelines, and governance AI Workflow Automation - DAG schedulers and model CI/CD Genetic Memory - Vector memory, knowledge graphs, and retention On-Device vs VM - Latency, privacy, and hardware trade-offs 💻 AI for Software Engineering Skills MySpec — Spec-driven development platform compiling guided developer discovery interviews into deterministic 4-file specification bundles ( constitution.md , requirements.md , solution.md , tasks.md ) with MCP server integration. AI Coding Foundations & Trust - AI coding interaction modes, human accountability, and safe adoption Context, Model & Prompt - Curate context, route tasks to suitable models, and request verifiable output Specification-Driven Development - Durable specs, API contracts, task decomposition, and validation Secure Model Selection, Privacy & AI Skills - Private evaluations, data controls, least privilege, and reusable workflows Multi-Agent Coding Workflows - Specialized roles, artifact handoffs, and human phase gates Model Context Protocol (MCP) - Safe, standard connectivity from AI clients to engineering tools AI PR Review Automation & Hooks - First-pass review, workflow triggers, and quality automation Agentic Systems Architecture & Frameworks - Governed agent platforms, BMAD Method, and Superpowers Working with Legacy Codebases - Map a safe change surface, preserve verified memory, and make small changes with evidence 🎓 Learning Resources Courses - AI engineering and solution architecture courses Certifications - AI engineering, solution architecture and cloud certifications 💼 Career Development Career Guide - Role snapshots, portfolio tips, and profile building 🎯 Interview Preparation Interview Prep - GenAI system design, practice questions, and frameworks 🏛️ Solution Architecture Solution Architecture Fundamentals - Core architecture principles, patterns, and practices Architectural Patterns & Styles - Design patterns and architectural styles for robust solutions Architecture Decision Records - Documenting architectural decisions and rationale Clean Architecture - Organizing code with dependency inversion and separation of concerns Business Architecture - Bridge business strategy and technology implementation Technical Architecture - Design robust, scalable, and maintainable technical solutions Requirements Analysis - Master requirements analysis and architecturally significant requirements Quality Attributes - Master quality attributes and their impact on architecture design Architecture Modeling - Master architecture modeling techniques and notations Architecture Governance - Establish effective governance frameworks for architectural consistency Architecture Assessment & Maturity - Evaluate and improve architectural capabilities Architecture Tools & Practices - Master tools, techniques, and practices for effective architecture Cloud Infrastructure - Master cloud infrastructure design and deployment Estimation Techniques - Master estimation techniques for accurate project planning Presales Architecture - Master presales architecture practices for winning proposals 🚀 Introduction Artificial Intelligence (AI) is reshaping how modern systems are designed. Solution architects are now expected to integrate machine learning models, data pipelines, and intelligent services into scalable and secure architectures that deliver business value. This repository curates learning resources, certifications, career tips, and interview prep materials to help you grow into an AI‑savvy solution architect or consultant. 🎯 What You'll Find Here Plain-language explanations of complex AI concepts Hands-on checklists and practical implementation steps Real-world examples with source attribution and "why it's awesome" explanations Visual diagrams using Mermaid for architectural clarity Curated resources from official docs, reputable sources, and "awesome" lists Career guidance for AI architects, consultants, and solution architects Interview preparation with system design frameworks and practice questions 🚦 Quick Start New to AI Architecture? Start with AI Architecture Patterns Choosing between models? Learn Model Routing Building RAG systems? Check Vector Stores & Embeddings Preparing for interviews? Jump to Interview Prep Career planning? Explore Career Guide 📖 How to Use This Guide ☕📚 Take a sip, learn a bit Grab a drink and enjoy the content and self-development. Each page follows a consistent structure: TL;DR - Simple explanation in plain language Quickstart - Actionable steps you can try now The Idea - Deeper context and concepts Key Concepts - Core ideas with real-world examples When to Use This - Clear guidance on applicability Real-World Examples - Practical implementations with sources Common Pitfalls - What to avoid and how Next Steps - Where to learn more 📄 License This project is licensed under the MIT License—see the LICENSE file for details. Made with ❤️ for the AI Architect community Contributions welcome! Feel free to open issues or submit pull requests.

10x your AI Architect journey with plain‑language, hands‑on guides that blend AI engineering and solution architecture.

Artificial Intelligence (AI) is reshaping how modern systems are designed. Solution architects are now expected to integrate machine learning models, data pipelines, and intelligent services into scalable and secure architectures that deliver business value.

This repository curates learning resources, certifications, career tips, and interview prep materials to help you grow into an AI‑savvy solution architect or consultant.

10x your AI Architect journey: plain‑language, guides that blend AI engineering + solution architecture.

5. Training Protocol, Convergence Behavior & Experimental Evaluation

💼 AI-Enabled B2B Invoice Management System AI-powered fintech application for invoice management, payment delay prediction, and workflow automation 🌐 Live Demo View Live App 🧠 Overview The AI-Enabled B2B Invoice Management System is a full-stack fintech project designed to streamline invoice operations and improve payment visibility through machine learning. This application helps businesses: manage invoice records efficiently predict expected payment delays identify late and high-risk invoices improve decision-making around accounts receivable workflows The frontend delivers an interactive invoice dashboard, while the backend ML pipeline processes invoice data and generates payment delay predictions using a trained Random Forest Regressor . ✨ Features 📊 Invoice Dashboard Single-page invoice management interface at / Dark-themed UI with glass-morphism styling Branded header with HighRadius and ABC logos Clean, responsive table-driven workflow 📄 Invoice Table Displays: Customer Name Customer Number Invoice Number Amount Due Date Predicted Payment Date Delay Badge Notes Additional UI behavior: USD-formatted amounts Alternating row backgrounds Hover effects Click-to-select rows Color-coded prediction states 🤖 AI-Powered Payment Prediction Predicts payment delay in days using a Random Forest Regressor Shows predicted payment dates directly in the dashboard Highlights: Late payments in red On-time payments in green Delay badge logic: 30+ days late → destructive red badge 1–30 days late → default delay badge On time / early → green badge ✏️ CRUD Operations Add Invoice Dialog-based invoice creation Fields: customer name, customer number, invoice ID, amount, due date, notes Validation for all required fields except notes Edit Invoice Enabled only when exactly one invoice is selected Allows editing of amount and notes Delete Invoice Supports bulk deletion Confirmation dialog before removal 🔍 Search & Filter Real-time search by: invoice ID customer name Automatically resets to page 1 on new search 📑 Pagination 10 invoices per page Previous / Next navigation Page counter “Showing X–Y of Z” indicator ✅ Selection System Individual row checkbox Select-all toggle for current page Selection state controls Edit/Delete button availability 🧠 ML Model Info Card Expandable model metadata card showing: Model: Random Forest Regressor R² Score: 0.4334 MAE: 24.99 days RMSE: 45.69 days Training / Test split: 10,687 / 4,581 samples Top 5 feature importances: Customer Identity Customer Number Invoice Amount Due Day Posting Day 🛣️ Routing / → Main invoice dashboard * → 404 Not Found page ✅ Testing Vitest setup for unit testing Playwright configuration for end-to-end testing 🏗️ Tech Stack Category Technologies Frontend React, TypeScript, Vite Styling Tailwind CSS, custom semantic tokens UI Components shadcn/ui State / Routing React Router Backend / ML Python ML Libraries scikit-learn, pandas, NumPy Model Random Forest Regressor Testing Vitest, Playwright Deployment Vercel 🧩 Architecture ┌─────────────────────────────┐ │ React + TypeScript │ │ Invoice Dashboard │ └──────────────┬──────────────┘ │ │ displays invoice data + │ predicted payment dates │ ┌──────────────▼──────────────┐ │ Demo Dataset / Sample UI │ │ 15 invoices with ML-fed │ │ predicted payment output │ └──────────────┬──────────────┘ │ │ generated from │ ┌──────────────▼──────────────┐ │ Python ML Pipeline │ │ data cleaning + features │ │ model training + scoring │ └──────────────┬──────────────┘ │ ┌──────────────▼──────────────┐ │ CSV Datasets │ │ NewDataset.csv / Archana │ └─────────────────────────────┘ 🤖 Machine Learning Pipeline The backend ML logic is implemented in: src/data/ml_pipeline.py Pipeline Responsibilities Loads invoice data from NewDataset.csv Cleans and preprocesses invoice records Converts CAD to USD Parses dates Handles missing values Performs feature engineering Encodes categorical customer fields Trains and compares multiple regression models Models Compared Linear Regression Random Forest Gradient Boosting Selected Best Model Random Forest Regressor Chosen based on best R² score Feature Engineering Day-of-month extraction from date fields Label encoding for: customer identity customer number Model Output Predicted delay in days Converted into predicted payment dates shown in the frontend 📁 Datasets public/NewDataset.csv Raw dataset containing 50K invoice records Original HighRadius invoice data source public/HRC71251W_Archana.csv Preprocessed feature dataset used for ML workflow 📸 Screenshots 📊 Invoice Management Dashboard 📊 Demo Data Note The frontend uses 15 sample invoices populated with ML-predicted payment dates . This serves as a demo dataset that bridges: the Python model output the React invoice dashboard This approach allows the UI to showcase prediction-driven workflows in a clean, interactive format. ⚙️ Installation & Setup 1. Clone the repository git clone https://github.com/archanav12321/B2B-APP.git cd B2B-APP 2. Move into the app directory cd b2bApp 3. Install frontend dependencies npm install 4. Run the development server npm run dev 5. Build for production npm run build 6. Preview production build npm run preview 🧪 Testing Run unit tests npm run test Run end-to-end tests npx playwright test 🚀 Future Improvements Connect the dashboard to live backend/API data instead of demo invoice records Add invoice upload and ingestion workflow Add analytics charts for payment trends and customer behavior Add risk scoring for overdue invoices Introduce authentication and role-based access control Deploy the ML pipeline as a prediction API Support downloadable reports and CSV export 👩‍💻 Author Archana Vellingiri Raja MS in Computer Science | Full-Stack Developer | AI/ML Enthusiast GitHub LinkedIn ⭐ Why This Project Stands Out This project demonstrates: real-world fintech problem solving full-stack frontend engineering with modern React tooling machine learning integration into business workflows practical prediction-based UI design clean product thinking, not just model training It combines software engineering + machine learning + business use case alignment , making it a strong portfolio project for roles in: Full-Stack Development Frontend Engineering Product Engineering AI-Enabled Application Development 📜 License This project is licensed under the MIT License.

AI-powered fintech application for invoice management, payment delay prediction, and workflow automation

The AI-Enabled B2B Invoice Management System is a full-stack fintech project designed to streamline invoice operations and improve payment visibility through machine learning.

The frontend delivers an interactive invoice dashboard, while the backend ML pipeline processes invoice data and generates payment delay predictions using a trained Random Forest Regressor .

┌─────────────────────────────┐ │ React + TypeScript │ │ Invoice Dashboard │ └──────────────┬──────────────┘ │ │ displays invoice data + │ predicted payment dates │ ┌──────────────▼──────────────┐ │ Demo Dataset / Sample UI │ │ 15 invoices with ML-fed │ │ predicted payment output │ └──────────────┬──────────────┘ │ │ generated from │ ┌──────────────▼──────────────┐ │ Python ML Pipeline │ │ data cleaning + features │ │ model training + scoring │ └──────────────┬──────────────┘ │ ┌──────────────▼──────────────┐ │ CSV Datasets │ │ NewDataset.csv / Archana │ └─────────────────────────────┘