
AI Engineer – Generative AI & Agentic AI
Accenture · Feb 2024 – Present
- Designed and built an AI-powered email response generator on Azure OpenAI (GPT-4), engineering prompt templates to classify incoming loan and credit card queries by intent and retrieve relevant account context from Azure Cosmos DB for grounded, compliant replies.
- Deployed the solution as a serverless Azure Function integrated with the internal ticketing system, and partnered with customer service leads to validate tone, accuracy, and edge cases before rollout - cutting average response turnaround time by 40% and easing workload for the support team.
- Built a customer-facing WhatsApp chatbot using Azure OpenAI, Azure Cognitive Search, the WhatsApp Business API, and Azure Functions, combining retrieval-augmented generation with multi-turn conversation flows and session/context management in Azure Cosmos DB to support 200+ intents, including balance inquiries, card block/unblock requests, and loan status checks.
- Collaborated with product and compliance stakeholders to define conversation guardrails, disclaimers, and fallback-to-agent handoff logic, then rolled out the chatbot in a phased regional launch - reducing inbound call center volume by 25%.
- Architected a multi-agent underwriting and fraud-screening system using CrewAI, defining specialized agents for document verification, credit risk scoring, and fraud-pattern detection with clear task boundaries and shared context.
- Implemented autonomous task-delegation and inter-agent communication logic, including escalation to human underwriters on low-confidence decisions, to balance automation speed with risk controls.
- Partnered with underwriting and risk teams to tune agent decision thresholds and validate outputs against historical cases cutting average loan-decision turnaround time by 30% while maintaining full audit traceability across all agent actions.
- Architected a multi-agent financial research copilot using CrewAI and LangChain, designing planner, retriever, and writer agents that autonomously query internal APIs and market data sources to support relationship managers preparing portfolio reviews.
- Implemented human-in-the-loop approval guardrails and citation tracking so advisors could verify every data point before client-facing use, ensuring compliance with internal research-disclosure policies.
- Collaborated with relationship managers and compliance stakeholders to validate briefing accuracy and tune retrieval relevance cutting portfolio-review prep time from hours to minutes.

