How RetaiLens AI Works
RetaiLens AI is an autonomous agentic operations platform—not a passive dashboard or a Q&A chatbot. Every perception triggers validated reasoning, memory retrieval, deterministic tool execution, and human-governed approvals.
End-to-End Execution Pipeline
Live aisle cameras (CAM-01..04), checkout lane video (CAM-CHK-01/02), and POS barcode scans.
FastAPI + OpenCV / YOLO interface extracts shelf occupancy %, bounding boxes, and confidence scores.
Agent ingests structured detection events, inventory deltas, and upcoming product expiry dates.
ReAct planner retrieves persistent PostgreSQL memory (velocity, lead-time drift, past approvals) before planning.
Server validates Zod tool arguments against RBAC & safety thresholds—never allowing raw LLM SQL.
High-value POs (>₹5,000), checkout overrides (<80% confidence), and discounts >15% pause for Store Manager review.
Idempotent transactional updates mutate stock, purchase orders, pricing, and GST-preview receipts.
Outcomes and human decisions persist to agent_memory across sessions for continuously smarter planning.
Executive Command Center, Event Timeline, Analytics, and Audit Logs update in real time.
Traditional Retail Software vs. RetaiLens AI
Traditional Retail Software
- Reactive — managers discover stockouts hours too late
- Manual monitoring across disconnected CCTV and POS tools
- Static threshold alerts with zero supplier lead-time context
- Manual workflows for PO creation and expiry discounting
- Limited historical memory for recurring SKU velocity surges
RetaiLens AI
- Proactive — detects low shelf occupancy and expiry risk early
- Automated CV + POS cross-verification at checkout lanes
- Context-aware ReAct reasoning with 13 validated backend tools
- Autonomous routine operations (verified sales, PO drafts, scans)
- Persistent PostgreSQL memory across inventory, supplier & pricing
- Human-controlled sensitive actions via Store Manager Approval Center