Rollie O’Malley

Real-time multi-agent AI companion · iOS App Store

• Production voice AI app built on the Gemini 2.5 Flash Live API
• 14 specialized tool handlers across 6 domains: kitchen, shopping, fix-it, restaurant, web search and memory
• WebSocket relay bridging client audio to Gemini’s native audio pipeline
• Firestore-backed persistent memory for context across sessions
• Real-time voice output with geolocation-aware weather context

Built with Gemini Live API, Node.js, WebSockets, Firebase/Firestore

AI email intelligence SaaS · live with real users

• Gmail API integration with a complete OAuth2 refresh-token flow
• One structured inference call per email for summarization, client detection, priority scoring (1–10) and calendar extraction
• Background job syncs every user every 30 minutes with incremental fetching

Built with Firebase Auth, Firestore, Docker, Caddy

Hardware · field-tested on a tree farm

Designed and built a drone that holds two gallon drums and sprays liquid.

• Designed the power system end to end
• Built ground-station control in place of a conventional RC link
• Debugged failures by bisecting the stack
• Real-world testing on a tree farm
• Future use: window cleaning for office buildings

24/7 automated phone receptionist for businesses

• Handles inbound phone calls with conversational AI
• Designed to scale with SMS/email confirmations, Google Calendar scheduling and daily summaries

Built with n8n, Vapi

MIT 16.412J Cognitive Robotics

• Simulated autonomous drone agent that picks up packages, avoids obstacles and manages battery constraints
• Implemented the decision-making logic and environment interaction

Built with Python

Machine learning for time-series forecasting

• Enter a stock ticker and the model predicts price movement over the next 50 days
• Monte Carlo sampling for uncertainty quantification

Built with Python

A 60-day case study on strength, body composition and joint pain in an adolescent male

Designed and ran a 60-day controlled experiment comparing high-volume and low-volume training. Low-volume protocols achieved equal or better outcomes with reduced injury risk.

• Formulated the hypothesis and controlled for nutrition, sleep and exercise selection
• Collected daily quantitative data across multiple metrics
• Identified the limits of bioelectrical impedance measurement and applied statistical corrections
• Wrote a formal paper covering methodology, results, discussion and practical applications

Varsity Robotics Team · first freshman ever selected

• 18.06Linear Algebra
  MIT OpenCourseWare
• 6.86xMachine Learning with Python: From Linear Models to Deep Learning
  MITx on edX
• 16.412JCognitive Robotics
  MIT OpenCourseWare
• UdacitySQL for Data Analysis · Intro to Algorithms · Data Wrangling with MongoDB
• AnthropicClaude Platform 101 · Claude Code 101 · Claude Code in Action · Introduction to MCP · MCP: Advanced Topics · Agent Skills · Subagents · AI Fluency · AI Capabilities and Limitations

Technical Skills

Languages

Python, Java, JavaScript

Frameworks & tools

PyTorch, Node.js, Express, React, Firebase/Firestore, Docker, WebSockets, OAuth2, n8n, Vapi, Git, CodeSandbox

APIs & models

Claude API, Gemini 2.5 Flash Live API, OpenRouter, Gmail API, Google Calendar API

Robotics & math

Odometry and field-centric control, kinematics and trigonometry, dynamic system modeling, applied linear algebra