The Problem Worth Solving
Picture this: A convenience store owner in Singapore wraps up another exhausting 14-hour day. It’s 10 PM, but before heading home, there are still 3-4 hours of administrative work ahead—manually tallying cash denominations, reconciling card transactions, entering invoice line items into spreadsheets, and trying to remember which products are running low.
This wasn’t an isolated case. This was the daily reality for thousands of small retail owners.
The hidden costs were staggering:
- 300+ products tracked manually in notebooks and fragmented spreadsheets
- 15-20 supplier invoices weekly, each requiring 15+ minutes of manual data entry
- Multiple payment methods (cash, NETS, cards, QR codes) reconciled by hand, with frequent $50-100 discrepancies
- Stockouts costing $800-1,200 weekly in lost sales due to guesswork-based ordering
- Missed early payment discounts worth 2-3% annually due to poor deadline tracking
The bottom line: Store owners were bleeding $2,000-3,000 monthly in operational inefficiencies while spending 25+ hours weekly on soul-crushing administrative tasks that added zero value to their business.
This was my client’s reality. And this is what I set out to solve.
The Solution: Sentinel
I built Sentinel—an AI-powered operations management system that doesn’t just digitize existing processes, but fundamentally reimagines how convenience stores operate. The goal was ambitious: automate 80% of administrative tasks while providing intelligent insights that small business owners could actually use.
What Makes Sentinel Different
1. Intuitive Web Interface
- 📤 Upload invoice photos directly through the web interface → AI extracts and validates all line items
- 📊 Submit daily closing figures via streamlined forms → system reconciles across all payment methods
- 🔔 Receive smart alerts for low stock, payment deadlines, and discrepancies
2. AI That Actually Works
Smart Reorder Engine: Analyzes 90+ days of sales patterns, factors in seasonal trends and supplier schedules, then generates optimal restocking recommendations with confidence scores
Invoice OCR Pipeline: Multi-stage extraction process achieves 95%+ accuracy even with poor photo quality or varied invoice formats
Anomaly Detection: Flags unusual transactions or inventory movements for review
3. Unified Financial Command Center
- 💳 Real-time reconciliation across cash, NETS (Singapore debit), credit cards, and QR payments
- 💰 Automated supplier payment tracking with early discount calculations
- ✅ One-click daily closing with discrepancy detection and audit trails
- 📤 Export-ready reports for accounting software integration
Upload your daily book closing screenshots from your POS,
AI does the job of tallying your daily book records and informs if there are any anomalies.
Technical Deep Dive
Architecture & Design Decisions
I architected Sentinel as a three-tier system optimized for both performance and maintainability:
- Flask over Django: Needed lightweight, API-first architecture with precise control over each endpoint
- SQLAlchemy ORM: Clean abstractions while maintaining raw SQL performance where needed
- SQLite over PostgreSQL (initially): Zero-configuration deployment, perfect for single-store MVP, trivial migration path to Postgres for multi-store scale
- Responsive Design: Built mobile-responsive from the ground up—works seamlessly on phones, tablets, and desktops
The AI Pipeline
Building accurate AI for messy real-world data required multiple specialized stages:
Invoice Processing Pipeline:
Photo Upload → Image Preprocessing → OCR Extraction → Context Validation → Product Catalog Matching → Confidence Scoring → Database Storage
Handling Real-World Complexity
- Challenge #1: Invoice Format Chaos
Flexible parsing engine using context-aware pattern matching, product catalog fuzzy matching, human-in-the-loop validation, and learning system - Challenge #2: Real-Time Data Synchronization
Event-driven architecture with database triggers, optimistic locking, transaction-based inventory adjustments, and WebSocket connections - Challenge #3: Non-Technical User Adoption
Intuitive interface with guided workflows, smart defaults, plain-language error messages, and comprehensive tooltips/documentation
Results That Matter
⏱️ Time Savings (The Human Impact)
- ⏱️ Daily admin time: 3.5 hours → 30 minutes (85% reduction)
- 📄 Invoice processing: 15 minutes → 2 minutes each (87% faster)
- 💰 Daily closing: 45 minutes → 8 minutes (82% faster)
Total: 22 hours recovered weekly—that’s nearly 3 full workdays back.
💰 Financial Impact (The Business Case)
- 💵 $2,500+ monthly savings in operational efficiency
- 📈 15% profit margin increase through inventory optimization
- 🎯 85% reduction in stockouts = $800-1,200 recovered weekly
- 💳 98% capture rate on early payment discounts (up from 30%)
Total: ROI achieved in 4 weeks
✅ Operational Excellence
- ✅ Data entry errors: 95% reduction
- ✅ Inventory accuracy: 70% → 96%
- ✅ Payment punctuality: 65% → 100%
- ✅ Order accuracy improved 40%, reducing waste
📊 Retail Operations – Dashboard
Client Impact
“Sentinel transformed my business. What used to take 4 hours every evening now takes 15 minutes. The AI recommendations help me order smarter—I haven’t had a major stockout in months. I’m capturing payment discounts I used to miss. This system paid for itself in the first month.”
— Store Owner, Daily Delights Convenience Store
Interested in building intelligent automation for your business? Let’s talk →
View on GitHub: github.com/mdavala/sentinel
Technologies: Python • Flask • AI/ML • Cloud Infrastructure • RESTful APIs • Database Design