Working on the low-latency scatter-gather ML architecture pioneered in https://arxiv.org/html/2403.13313v1 for a high-quality, realtime speech application in healthcare

Led development of the medical-core of the conversational AI systems:

• - Built medical safety engines for adverse event detection, medication reconciliation, and emergency (911) call handling with graceful escalation

• - Architected multilingual voice AI infrastructure supporting real-time language detection, ASR switching, and TTS fallback across

• - Designed a patient scheduling system with appointment booking, timezone localization, and provider integration APIs

• - Developed a transfer protocol framework enabling seamless handoffs between AI and human agents with consent workflows

• - Drove reliability by improving non-technical input validation, implementing concurrency controls in eval tools, and common-sense fallbacks

• - Improved latency by identifying poor parallelism, saving 50ms per turn

• - Improved accuracy/ automated eval frameworks, getting us that second 9 in many of our supporting LLMs

• - Saved $1M/year by noticing and fixing a poor implementation detail from the prototype days that nobody spotted

Cross-Platform(QT/ C++ Server), Full Stack (.Net/React), and AI/ML Implementation Engineer

ML:

• - Architected and led the development of a Github CoPilot-like experience for Roblox's IDE end to end, including LSP changes, microservice backend development, and integration with OpenAI's API

• - Collaborated on AI success metrics (edit distance, % code generated, etc)

• - Implemented a basic RAG pipeline for Q&A

• - Implemented POC of agentic system with Langchain for a building-productivity focused chatbot

• - Implemented experiment methodology and pipeline for AB testing many features in parallel with proper isolating.

• - Fine-tuned model for script performance (LoRA)

• - Used open source models from hugging face for toxicity filtering (BERT)

Cloud:

• - Built Packages (a Roblox way of sharing non-trivial assets) from zero to one. Implemented Zero UX designs for synchronization. Improved storage size of games by symbolic linking, reducing S3 storage costs and enabling reuse and versioning.

• - Improved/ encouraged Cloud-based development, implementing strategies for new Developers to think of their game as always-in-the-cloud

• - Migrate monolith components into micro-services

• - Implemented User Sandboxing with CURL

Performance-Related:

• - Made the IDE faster and more responsive by identifying unnecessary UI locking/ bottlenecks and developing a strategy for resolving/ preventing these issues in the future. Cumulatively saved 10,000+ hours per year by saving 10 minutes per day from every user.

• - Improved the output window, making it faster & more performant with multi-threading, batching, and selective rerendering. Logging in large games was infeasible before this.

• - Rearchitected the Find Replace widget, improved performance with multi-threading and parallelization. Performs better than VSCode by utilizing extensive caching and indexing

• - Ground-up rebuild of Studio's debugger, implementing DAP and MVC architecture.

August - October - Java , SDET

• Architect scalable tools and methodologies for Cucumber/ Gherkin

• Automate VSphere/VMWare deployment for performance testing

July - August- C++, Systems/ performance

• Use VS2015 performance profiler to identify/ fix bottle necks

• Address concerns with scaling current implementations for increased loads on server

March - June- C#, Team Lead

• Lead small (< $1,000,000) contract to completion

• Establish baselines, needs, and means of execution

• Handle vague and often misunderstood requirements/ requests

• Set up build and release environments

• Handle issues with third party hardware integration software issues, including meeting developers in constructive, efficient manner

February -March - C#, Application development

• C# development using WPF

• Handle async concurrency problems

• Update legacy client to use new back-end architecture

• gRPC client interface

June - February - C++, Application development

• Updated legacy application/client to touch-optimized UI using QT

• Updated military communication software to meet new standards

• Created well designed and documented DLLs for new protocols/message sets

• Handled performance/ multi-threading issues and bottlenecks

• Develop eye tracking application for lab use

• Integrate with motion sensors

• Convert pre-existing C application to more modern, open-source framework

• Technologies: Python, PsychoPy

• Independent developer on a project from an external client

• Navigated difficult cost/time estimation to meet needs of both customer and employer

• Technologies used: Swift, OAuth