Hi, I'm Mario Dias.

Navi Mumbai, India

• Built a real-time audio-based RAG assistant for the CEO, CFO, and Investor Relations team to generate context-aware, low-latency talking points during quarterly earnings calls using Azure Speech, Azure OpenAI, FAISS, and FastAPI.
• Developed ampliFi Data Integrity, a configurable machine learning-based anomaly detection system for financial datasets to early detect and prevent data quality issues, using SynapseML, Databricks and MLFlow.
• Deployed an autonomous agent for financial signoff summarization as part of the UBS ampliFi initiative, leveraging LangGraph, FastAPI, and Azure OpenAI to streamline and assist controllers.
• Won UBS India Hackathon 2024 by developing a multi-agent RAG-based AI assistant using LangChain, LangGraph, Azure OpenAI, and Azure AI Search to provide application-specific user assistance and automated incident reporting, showcasing innovation in enterprise AI.
• Designed and optimised an ETL pipeline in Databricks ingesting millions of records for daily balance-sheet reporting, using a concurrent queue to process data from 13+ regions across 100+ daily jobs.
• Boosted the performance of critical Databricks jobs from 45 minutes to under 8 minutes, significantly enhancing data availability, reducing costs, and improving operational efficiency.
• Played a key role in the development of an efficient Automation Testing framework for the application, leveraging Spring and Cucumber to encourage Behaviour-Driven Development and reduce testing time, leading to increased release cycles.
• Recognised as a UBS Certified Gold Engineer for technical excellence and consistent delivery.
• Core member of the UBS AI Ventures Challenge, mentoring graduates on AI and innovation against real-world business problem statements.

Mumbai, India

• Managed and updated website content, redesigned layouts for events, and optimised performance with plugins and caching.
• Collaborated with the team on timely updates and technical issue resolution.

Greater Noida, India

• Developed and trained an end-to-end deep learning model for activity recognition to detect suspicious behaviour in surveillance footage.