AegisFlow (InvoiceIQ) ⚡

AI-Powered Financial Intelligence & Risk Management SaaS

AegisFlow is a full-stack, enterprise-grade SaaS platform designed to bring advanced machine learning and predictive financial intelligence to the Pakistani business market. It moves beyond standard invoice tracking by utilizing AI to cluster client risk, forecast liquidity, and stress-test cash flow against macroeconomic shocks.

🚀 Tech Stack & Architecture

Frontend (The Interface)

• Framework: Next.js (React)
• Styling: Tailwind CSS with a custom Dark Mode Glassmorphism aesthetic.
• Data Visualization: Recharts (Cash Flow & Predictive Trajectories).
• Localization: Fully localized for Pakistani Rupees (PKR).
• Hosting/Deployment: Vercel.
• Monitoring: Vercel Web Analytics (@vercel/analytics).

Backend (The Logic Engine)

• Framework: Python / FastAPI.
• Hosting/Deployment: Railway.
• Machine Learning & AI:
  • K-Means Clustering (scikit-learn): Analyzes historical payment delays and invoice volumes to assign mathematical Risk Tiers (High, Medium, Low) to clients.
  • LSTM Neural Networks (PyTorch): Long Short-Term Memory models to project 30, 60, and 90-day cash flow liquidity.
  • Generative Adversarial Networks (GANs): Simulates worst-case economic scenarios (Market Constriction, Hyper-Inflation) to stress-test financial survivability.

Database & Authentication (The Data Grid)

• Database: Supabase (PostgreSQL).
• Security: Strict Row Level Security (RLS) policies.
• Authentication: Supabase Auth (Magic Links / Email Verification) locked to production routing.

---

🛠️ The Engineering Journey & Development Phases

The development of AegisFlow was executed in four distinct engineering phases, overcoming significant architectural and state-management challenges.

Phase 1: Foundation & The UI Grid

The project began by establishing a secure, scalable data grid. We integrated Next.js with Supabase, setting up the relational tables for clients and invoices. The UI was engineered with a premium Glassmorphism design system, allowing users to generate branded PDF invoices and manage client profiles.

Phase 2: The AI Processing Pipeline

The core differentiator of AegisFlow is its Python-driven intelligence. We stood up a FastAPI backend on Railway to handle heavy mathematical computations that a Node.js server shouldn't process. We successfully linked the Next.js frontend to the Python API, allowing the platform to pass live database metrics into the K-Means and LSTM models.

Phase 3: System Debugging & Circuit Optimization

This phase involved closing critical logical and routing loops:

• The 422 Schema Sync: Resolved a massive 422 Unprocessable Content error by rigidly aligning the Next.js JSON payload with the Python Pydantic models, explicitly mapping variables to Number() to ensure clean data transfer.
• The Recharts Render Bug: Eradicated the classic width(-1) negative dimension rendering error by injecting explicit minWidth={1} DOM boundaries, preventing race conditions during the initial React layout pass.
• The Dynamic Time-Sync: Addressed a critical logical flaw where the static database lacked a clock. We engineered a frontend "Time-Sync" circuit that dynamically compares due_date against the live calendar date ($\Delta t$), visually flagging invoices as Overdue and recalculating the precise avg_payment_delay_days before sending the data to the AI.

Phase 4: Production & Market Readiness

To transition from a developer environment to a live product:

• Production Authentication: Rewired the Supabase Site URLs and Redirect wildcards (/**) to seamlessly route email verifications to the live Vercel domain instead of localhost.
• Telemetry & Analytics: Injected Vercel Web Analytics into the root layout to monitor live traffic and user routing.
• Native Feedback Loop: Deployed an in-app feedback PostgreSQL table and UI modal, allowing beta testers to communicate directly with the database.

---

🔮 Future Roadmap

• Map the production Vercel build to a dedicated .tech domain.
• Expand the GAN stress-testing parameters to include industry-specific regional shocks.
• Automate the K-Means clustering via Supabase Edge Functions (CRON jobs) for background processing.

Developed in Pakistan.