Software Engineer Intern at CodeGen International (Pvt) Ltd

Architected core backend components of the TravelBox AI Assistant, a multi-agent AI system translating natural language into complex post-booking operations for travel support teams. Led migration from LangChain to LangGraph by redesigning orchestration from sequential agent execution to graph-based workflows for autonomous task-specific agents that safely execute multi-step booking operations with dynamic planning, validation, and user confirmation. Engineered a reusable object-oriented framework for building AI agents, including WebSocket progress streaming, token usage monitoring, latency tracking, and execution telemetry for observability. Optimized determinism and reasoning by offloading presentation logic to backend modules and implementing structured output validation, context injection, and execution constraints to improve reliability, reduce latency and token usage, and mitigate LLM failure modes.