Work

Leading the AI engineering stream for live customer voice calls and digital chat for all HSBC markets globally. Built and scaled event-driven pipelines to support 600 concurrent live conversations, scaling up to 1,500 requests per second at peak loads; streaming real-time AI context to cut HSBC customer resolution times from 15 minutes down to 5 minutes.

An in-house LLM intelligence engine that replaces rigid flowcharts with natural-language journey contracts. Connects approved Model Context Protocol (MCP) tools for transactional actions, pairing autonomous reasoning with code-enforced consent and safety gates across live voice and chat.

An orchestrated multi-agent platform where specialized AI agents deliberate and debate amongst themselves to stress-test decisions and drive meetings to conclusion. Scaled to 300 concurrent live simulations; piloted in hands-on simulation exercises with global CXOs at the HSBC Global Executive Leadership OPCO in Nansha, China and the India OPCO in Mumbai.

An interactive speech-to-speech roleplay environment where simulated customer personas rehearse realistic scenarios with human agents, coupled with automated scoring on compliance and customer service quality.

Built a specialized LLM data assistant for Commercial Bank Relationship Managers using a fine-tuned StarCoder-15B model, delivering precise financial answers while keeping hallucination under 12%.

Architected zero-downtime deployment workflows and advanced report-rendering pipelines for institutional financial reporting systems, handling massive daily transactional data.