The Bambu technology platform creates high-EQ conversational AI assistants tailored to the healthcare industry. Customer-facing offerings include “Willow,” which supports licensed therapists and their clients between sessions (available on the Apple App and Google Play stores), an assistant that matches prospective patients with providers, and a “coach” for GLP-1 patients.• Perform fine-tuning and RAG prompt engineering for a constellation of LLM AI models on the server side, utilizing hosted AI services from OpenAI and publicly available models (including Llama 2 and Llama 3.1); use PyTorch to fine-tune Llama models using cloud GPU servers. • Implemented a C#/.NET server for Willow that provisions the app with secure access to AI models and provides APIs to interact with databases and data feeds for RAG. Maintain a reliable Kubernetes deployment on AWS for mobile that is highly available, scalable, secure, and cost-effective. • Analyze LLM behavior and log probabilities to guide content team in developing purpose-built, sculpted personalities. Establish testing and maintenance processes for the fine-tuning corpus to prevent personality drift between releases. Memory system database injects contextually relevant memories of previous conversations and goals, facilitating fresh, relevant interactions and analysis with each new session.

Designed and implemented the server-side game engine for Kingdom Maker, an immersive MMO with real-time PvP combat. In addition to my role as Principal Architect, I owned the 24/7 prod environment, managing a team of SREs. The game is deployed and managed on Kubernetes (AWS EKS); architecture includes APIs for mobile clients, a gRPC service mesh for backend communications, and observability using Prometheus.• Game engine supported acquisition of up to 200k new users each day in a highly concurrent, fully connected MMO experience. • Implemented server engine to support very low latency (below 100 microseconds) distributed database transactions at high scale. This included writing a distributed transaction manager for Redis and MySQL as well as a distributed cache manager between server replicas and clients. • Retooled the architecture post-worldwide launch to operate at web scale with a more cost-effective Kubernetes deployment. • Designed and implemented a highly reliable distributed backup and restore system for prod to maintain multiple cloned staging and dev deployments on separate Kubernetes clusters.• Performed zero-downtime upgrades to game servers and clients multiple times per week.

Led Sysdig’s product transformation into a modern Prometheus-based SaaS monitoring solution. • Directed and architected development of Sysdig’s SaaS system security and observability platform. • Managed multi-platform global team, leading by example with hands-on implementation.

Developed overall technical strategy for the company’s video¬ and music sharing app, available on iOS/Android/desktop. Managed team of server engineers and SREs in California and Europe and shared responsibility for the Data Science team. Up to 30 million monthly users recorded, edited, and shared music videos both on and off the Triller platform, requiring a highly available and scalable server to handle high transaction rates and bandwidth. • Rebuilt existing server infrastructure in-flight to support a user growth rate that doubled every two months, maintaining 99.99% availability while responding to DDOS and coordinated Bot attacks.• Reduced cost of per-user hosting by more than 90% by transitioning server stack from AWS EC2-deployed servers to Kubernetes clusters running containerized microservices and shifting performance-sensitive endpoints from Python to Go.• Developed ML, Convolutional Neural Network (CNN) models using TensorFlow/Keras that classified key frames of uploaded videos to identify content violating terms of service. Developed internal tools to facilitate human-in-the-loop training for CNNs.• Ensured seamless integration with Triller partners, including Apple Music, Spotify, Twitter, Instagram, YouTube, Facebook, and major record labels.

Member of Architect Core Team evaluating and implementing Autodesk’s current and future state enterprise landscape.• Architected and optimized customer support platform during Autodesk’s transition to a subscription/consumption model, enabling millions of daily users to access seven million unique pieces of content such as interactive media, product manuals, videos, and forums.• Researched and developed AI models for automated customer support. Used Word2Vec and SpaCY to build models of existing support content and implemented Approximate Nearest Neighbor searches (with “ANNOY”) for retrieval. Researched application of LSTM models for the same purpose.• Served as architect on content team for conversational bots built with IBM Watson Conversation.• Oversaw technical strategy for customer, partner, and employee knowledge base and technical support tools. Technology stack: Lucene/SOLR, AWS, Salesforce, and ML and data science toolkits in Python, Scikit-learn, and SpaCy.

Led the technical effort for KIXEYE’s next generation of MMO game servers and services; managed architecture and implementation of backend servers across multiple product lines.• Delivered significant improvement in concurrent performance and scale: developed AWS cloud deployment, with servers written in Node.js using Redis, Aurora, and S3 for persistence. This breakthrough system employed dynamic scaling to support hundreds of thousands of simultaneous players, automated testing tools, low cost of development, and extremely low cost of operation per player. • Implemented backend game engine for VR games using Unreal Engine 4 and C++. Developed and deployed “Armor,” a Steam tank combat game.• Created scalable, low-latency game simulation engine in Node.js for an MMO game with Unity clients.• Developed server-side AI for NPC game opponents within the MMO, using pathfinding algorithms such as A* and minimax algorithms with alpha/beta pruning.

Architected and implemented server layer for the Emotive PaaS, which provided development and deployment tools for iOS, Android, and Windows 8 apps, and hosted server components of customers’ mobile apps on AWS and Azure. Customers provided their server components using Node.js, with interfaces to popular cloud services such as Salesforce.com and LinkedIn.com.

• Designed and implemented server layer for Pokemon online trading card game, including scalable database layer for "digital object” trading of online card entitlements. System handled billions of virtual card entitlements, with transactional player-to-player trading of cards.• Implemented auction clearinghouse for card trading, supporting a community of hundreds of thousands of players actively trading cards.• Implemented scalable SOA for clustered servers using Java Spring, Fuse ESB, and PostgreSQL to support a target audience of hundreds of thousands of concurrent players.

• Designed proteomics system for identifying PTM’s (post-translational modifications of proteins) based on human genome; built algorithms for in silico digestion of peptides and matching mass spectrometer peaks.• Developed Java algorithm for accurate peptide identification based on data from tandem mass spectrometers which significantly out-performed competitive products.• Developed scalable peptide and protein search infrastructure for cloud deployment on AWS. System utilized MongoDB for a large clustered database of peptides derived from the human genome.

Played a key role in post-acquisition transition, enabling successful incorporation of AmberPoint’s technology into Oracle’s SOA management product lines.

• Architected AmberPoint's suite of tools for management and monitoring of SOA applications. Directed both internal SOA components and end-user experience.• Worked with other founders to set initial product strategy and technical direction and led new product initiatives through the life of the company.• Developed SOA governance tools with Grails services and Flex UI.• Optimized AmberPoint’s Java SOA infrastructure to manage transaction rates of up to several thousand transactions per second on each node. Worked closely with Fortune 50 customers to ensure that products scaled to their enterprise requirements.• Patents: Apparatus and method for content and context processing of web service traffic (patent #7,549,153); apparatus and method for web service message correlation (patent #7,743,150).

• Managed XML tools development. Led team that developed XSLT-based web UI generationtools.• Led team to incorporate J2EE support into Netbeans IDE.• Served on JCP committees for Servlet 2.3 and JSF. Worked with Java architects to enhance andevolve J2EE.• Frequent technical presenter at various conferences, including Sun’s JavaOne.

• Acted as architect of Forte tools for distributed application development. Led teams that designed and developed Forte IDE 1.0 product and created Forte’s award-winning OML tools for model-based application development.• Served as architect and team leader for Forte's J2EE development tools, which were the primary target for Sun's acquisition of Forte.