Associate Director, Data Scientist at S&P Global

I build practical AI tools for enterprise workflows, with a focus on generative AI, agents, and data science. My background spans Python development, software engineering, machine learning, and deep learning.

Here on GitHub, I explore AI engineering, AI architectures, AI evaluations, applied AI, and the Model Context Protocol (MCP) — from real-time voice agents and tool integrations to reproducible benchmarks.

LinkedIn · Explore my repositories

Real-time voice agents

A framework for building voice agents with Gemini Live API, knowledge retrieval through Gemini File Search, and custom tools. Supports local microphone conversations and phone-call integration through Twilio.

Evaluating how agents discover tools

A benchmark comparing keyword search, embeddings, decision models, and LLM agents on MCP tool discovery. Evaluates tool selection, abstention, latency, cost, and calibration against shared catalogs and held-out questions.

Building blocks for tool integrations

Minimal Python and TypeScript MCP server templates for local stdio and remote Streamable HTTP transports. Demonstrates transport setup, tool registration, typed inputs, and request/response flows.

Comparing structured decision models with LLMs

A runnable Python benchmark for support-ticket routing, comparing TypeSafe Jev with an LLM using the same tickets and department labels. Includes accuracy, latency, and estimated-cost reporting, plus a credential-free dry run.

Agent-assisted cloud operations

CLI runbooks and agent skills for setting up, operating, troubleshooting, and tearing down Oracle Cloud infrastructure. Includes pre-flight allowance checks and step-by-step guidance for network and compute resources.

• AI engineering: building agents, integrating models and tools, and developing reliable AI applications.
• AI architectures: designing voice agents, retrieval workflows, and systems that connect models with external services.
• AI evaluations: reproducible comparisons of accuracy, latency, cost, calibration, and tool selection.
• Applied AI: turning models into useful enterprise workflows and developer tools.
• MCP: tool discovery, server design, transport patterns, and interoperable agent integrations.

Find me on LinkedIn to discuss AI engineering, architectures, evaluations, applied AI, and MCP.