EnoXAI - Enorsia Customer Support Agent

I personally architected and built the entire stack — from the LangGraph agent graph design and tool node implementation to the CLIP-FAISS indexing pipeline, Laravel internal API integration, and production deployment infrastructure.

Key decisions I made:

• LangGraph over single-prompt chains — customer support flows require complex branching logic (order lookup → verify status → check cancellation eligibility → process or escalate). A graph with explicit tool nodes provides full observability, prevents the agent from skipping validation steps, and enables graceful error recovery.
• CLIP + FAISS for visual product search — customers frequently describe products visually ("blue floral maxi dress with ruffles"). CLIP embeddings enable accurate visual similarity matching across the entire catalog without relying on potentially incomplete text metadata, dramatically improving product discovery accuracy.
• Comprehensive tool node architecture — designed 15+ specialized tool nodes covering the complete customer journey: order status, cancellation requests, delivery incident checks, invoice resends, address updates, return creation, return status checks, coupon validation, support ticket creation, live product search, and more — each with strict validation and context-aware execution.
• FastAPI as the agent runtime — async-native architecture handles concurrent conversations efficiently, integrates seamlessly with Python ML tooling (CLIP, FAISS), and maintains clean separation from the Laravel monolith that owns authoritative business logic and transaction state.
• Structured knowledge base coverage — designed knowledge retrieval across 25+ customer journey touchpoints covering accounts, checkout, delivery, payments, returns, store policies, B2B, and more — ensuring the agent can answer any question with consistent, accurate information drawn from Enorsia's official knowledge base.

The agent handles the complete customer journey without human intervention: answering product questions, suggesting alternatives, checking order status, processing cancellation requests, managing returns, validating coupons, updating delivery addresses, resending invoices, checking for delivery incidents, and escalating complex issues to support tickets with full context.

Real-time actions the agent automates:

• 📦 Order Status & Details — Look up live order status and provide complete order information instantly, including items, quantities, shipping details, and delivery estimates.
• ❌ Cancellation Requests — Accept and process order cancellation requests directly within the chat flow with eligibility validation and immediate feedback.
• 🚨 Delivery Incident Check — Identify orders flagged with delivery incidents and surface them proactively to customers with clear explanations and next steps.
• 🧾 Invoice Resend — Resend lost or missing invoices to the customer's registered email on request, maintaining customer satisfaction without agent involvement.
• 📍 Address Lookup & Update — Show current shipping address and update it upon request where order status permits, with real-time validation from Laravel.
• ↩️ Return Management — Create new return requests and check the status of existing ones fully automated, including eligibility verification and return label generation.
• 🏷️ Coupon Validation — Validate discount and coupon codes in real time with accurate application rules, so customers know exactly what they're getting before checkout.
• 🎫 Support Ticket Creation — Escalate complex issues by raising support tickets directly, automatically passing full conversation context to human agents so customers never need to repeat themselves.
• 🔍 Live Product Search — Search the Enorsia product catalogue in real time based on customer descriptions, with category links and visual matching for accurate discovery.

Business impact:

The AI Support Agent transforms Enorsia's customer service operations by delivering enterprise-grade automation without compromising quality or accuracy.

• 24/7 Customer Support — Customers get accurate help at any hour with no wait times, no out-of-hours gaps, and no staffing costs for overnight shifts. Response availability increases from business hours to 24/7/365.
• Faster Response Times — Instant answers to common questions mean customers spend less time waiting and more time shopping, directly improving conversion rates and customer satisfaction scores.
• Reduced Agent Workload — 15+ routine support flows are handled automatically, freeing support staff to focus on complex cases that actually need human judgment, empathy, and escalation handling.
• Consistent & Accurate — Every response is drawn from the same structured knowledge base and validated through the agent graph's explicit tool nodes — no misinformation, no off-script answers, no hallucinated product details.
• Peak-Season Scalable — Handle Black Friday or holiday surges without adding headcount. The agent scales to any volume instantly with horizontal deployment, maintaining performance across thousands of concurrent conversations.
• Seamless Escalation — When a case needs human attention, the agent raises a support ticket with full conversation context automatically — no customer needs to repeat themselves, ensuring smooth handoffs and maintaining continuity of care.