Principal Engineer, Azure Observability at Microsoft

Used time series ML models to detect anomalies and developed MCP Servers to enhance Agentic AI offerings in Azure Observability. Focused on building scalable backend services leveraging AI models for enhanced monitoring and anomaly detection.

Founder at Neoba AI

Created an MVP to solve ChatGPT for the enterprise by building crawler tools to download content (text and tables), compute embeddings, and store them in VectorDBs like Pinecone and ChromaDB locally. Developed custom prompts integrated with OpenAI and locally hosted LLMs (Llama, Mixtral) to answer user queries effectively.

Senior Staff ML Engineer at Twitter (X Corp)

Part of the team building a supercomputer cluster for generative AI (Twitter 2.0). Served as tech lead for ML serving, delivering a highly available, high scale (60+ million requests/second), low latency (p999 <5ms) ML serving solution used by diverse ML applications including recommender systems and NLP media services.

Staff Engineer at Twitter

Tech lead for Twitter's large object store (~100 petabytes designed for exabyte scale) supporting 100x millisecond latency, ~500k QPS with high availability, durability (4x copies), hybrid storage cost optimized across Twitter, Google GCS, and Amazon S3. Architected and implemented the video transcoding backend, the largest distributed computing service at Twitter, delivering the service ahead of schedule and hardware budget. Designed and implemented Android photo features including smart cropping based on face location and dynamic photo experience optimization on low bandwidth mobile networks.

Architect & CTO, Engineering Services at Zynga

Reduced developer startup time for new games from 2 weeks to 15 minutes enabling rapid game prototyping. Delivered scalable game deployment system capable of updating software on thousands of servers in minutes allowing multiple daily game updates with minimal user disruption. Created a distributed shell for executing commands interactively on thousands of machines with sub-second latency for debugging and operational tools. Optimized storage CAPEX/OPEX and engineering costs by evaluating HDFS/HBase, Cassandra, Redis, MongoDB, Membase.

Lead Software Design Engineer, Bing Search at Microsoft

Architected and implemented the web crawler and indexer for Bing Search capable of competing with Google at scale by processing petabytes of data daily. Built and led a high performance team of 10 engineers to deliver a brand new web crawler and indexer for Bing within 12 months.

Lead Software Engineer, Enterprise Search at Microsoft

Owned Search Application Platform and extensibility. Opened up Search Indexing Pipeline enabling scenarios not possible with Google Appliance. Led team building a full ship cycle in Office 12 responsible for crawler and data gathering components. Delivered relevance comparable to Google Appliance and 10x crawl scale improvement. Delivered full text search for SQL 2005 and Exchange 2007 with 400x scale and 18x crawl speed improvement. Delivered text extraction features for Windows Vista desktop search.

Software Design Engineer, Enterprise Search at Microsoft

Integrated full text search for SQL Server by designing and implementing out of process work execution, ported MS Search code into SQL Server threading, exception, memory, and asynchronous IO models. Made architectural contributions for Search platform in SQL Server. Designed and implemented Thesaurus using Lex/Yacc based rule parser, secure component plug-ins, and deadlock detection framework. Created static source code analyzer to find destructors with unhandled exceptions. Part of the team that scaled MS Search to 2 billion documents.

Software Design Engineer, C# and .Net Runtime at Microsoft

Designed and implemented .Net verifier for x86 JIT compiler and contributed substantially to verifier and .Net Intermediate Language specification. Made contributions to the definition of the C# language. Designed and implemented isolated storage for .Net Frameworks Library, a sandbox file system for managed components. Part of a three-member team that implemented declarative security, security policy, and stack walking for code access security.