● Designed enterprise multi-tenant SaaS platforms using LangChain and LangGraph, selecting multi-agent architectures only when the problem required durable state across multiple reasoning steps. ● Built LangGraph stateful workflows for multi-agent systems with conditional routing and intermediate checkpoints for complex reasoning chains. ● Designed RAG pipelines using pgvector (for PostgreSQL workloads with relational joins) and Qdrant (for high-throughput standalone semantic search). ● Developed a production AI chatbot achieving sub-200ms response latency using Redis semantic caching and structured prompt templates with per-session context scoping. ● Integrated Salesforce, HubSpot, Stripe, and Twilio via OAuth2 with retry logic and idempotency handling to prevent silent failures in enterprise workflows. ● Engineered microservices architecture with gRPC for inter-service communication on AWS EC2; used REST for external client-facing APIs. ● Implemented async task pipelines with Celery and RabbitMQ to decouple heavy data processing from the request path with durable queuing. ● Built Redis caching layer for session management and API response optimisation, with TTLs calibrated to actual data change frequency. ● Delivered AI-Powered Multi-Subject Tutoring Backend: 10-agent LangGraph pipeline with subject-specific RAG (pgvector + OpenAI embeddings) and full LLM observability via Langfuse. ● Built Enterprise Data Intelligence Platform: LangGraph agentic workflows integrating Salesforce, HubSpot, cloud storage, and databases with per-client RBAC and NLP pipelines for entity extraction and sentiment analysis. ● Created Healthcare Patient Assistance Chatbot: RAG on Qdrant grounded in clinical guidelines, multi-turn per-session memory, end-to-end encrypted data for HIPAA alignment. ● Built Marketing Analytics Dashboard: Scrapy ETL pipelines normalising metrics from Google Analytics, Facebook Ads, Salesforce, and HubSpot into unified KPIs with Celery-scheduled automated reporting.