FinTech SaaS and Predictive Risk Intelligence
AegisFlow (InvoiceIQ)
AI-Powered Financial Intelligence and Risk Management SaaS
AegisFlow is live as an enterprise FinTech SaaS platform delivering AI-powered risk intelligence to daily finance operations. Teams use it to classify client risk, forecast 30, 60 and 90 day liquidity, and stress-test cash flow under adverse market conditions with production-ready reliability.
Challenge
Finance teams needed one platform that combines enterprise reliability, audit-safe operations and advanced ML forecasts without slowing down day-to-day decisions.
Solution
We delivered a production split-architecture: Next.js SaaS frontend for finance operators, Python FastAPI intelligence services for heavy ML compute, and a Supabase security/data layer with strict Row Level Security. The result is a fast and trustworthy financial intelligence cockpit already used in live workflows.
Tech Stack and Architecture
Frontend
- • Next.js and React for fast SaaS UI
- • Tailwind CSS with premium glassmorphism visual system
- • Recharts for liquidity and trajectory visualization
- • PKR-first localization and production deployment on Vercel
Backend and AI
- • Python and FastAPI services deployed on Railway
- • K-Means clustering for mathematical risk tiers
- • LSTM models for 30, 60 and 90 day liquidity forecasting
- • GAN simulation for macro-shock stress testing
Data and Security
- • Supabase PostgreSQL as core transactional data grid
- • Strict Row Level Security for tenant-safe access
- • Supabase Auth with production-safe magic link routing
Engineering Phases
Phase 1: Foundation and UI Grid
Established relational data structures for clients and invoices, then delivered a premium operator UI with branded invoice generation and profile workflows.
Phase 2: AI Processing Pipeline
Connected the Next.js product layer with a dedicated Python intelligence API so live database metrics could feed K-Means and LSTM models in real time.
Phase 3: System Debugging and Optimization
Resolved schema transfer failures, chart rendering race conditions and static date logic by aligning payload contracts, forcing safe chart dimensions, and engineering a live time-sync circuit.
Phase 4: Production Readiness
Switched authentication redirects to production domain routing, added telemetry through Vercel Analytics and launched an in-app feedback loop wired directly to PostgreSQL.
Debugging Highlights
- • 422 schema sync issue fixed by strict JSON-to-Pydantic contract alignment and explicit numeric casting.
- • Recharts negative width bug removed via enforced minimum render boundaries.
- • Dynamic time-sync logic added to compute overdue status and payment-delay metrics against live dates.
Roadmap
- • Expand GAN stress simulations for industry-specific regional shocks.
- • Automate K-Means clustering with scheduled Supabase Edge Functions.