HARSH PATEL
Senior Software Engineer • AI Agents • Cloud Infrastructure
Building scalable AI agent platforms and cloud infrastructure at Oracle. Architecting enterprise-grade solutions that combine Gen-AI technologies with robust distributed systems, delivering measurable improvements in performance, security, and compliance.
ABOUT_ME
🎓 Education
University at Buffalo
MS in Computer Science
Machine Learning & Distributed Systems
2021 - 2022
💼 Current Role
Oracle Corp.
Senior Member of Technical Staff
AI Agents & Cloud Infrastructure
Redwood Shores, CA
⚡ Expertise
Core Focus
AI Agent Orchestration
Cloud Infrastructure
Distributed Systems
MEASURABLE_IMPACT
WORK_I'M_PROUD_OF
Edit-Anything: Interactive Image Editor
Won 2nd place at Google DeepMind's Nano Banana Hackathon in San Francisco. Built interactive image editing tool allowing users to seamlessly edit images by selecting and editing text directly in-place on images.
OSWorld-G: GUI Grounding Benchmark
Built computer-use grounding benchmark and dataset (Jedi) with 4M training examples for training GUI agents. Achieved state-of-the-art performance in accurately locating and interacting with UI elements across platforms.
OmniParser: Screen Parsing for AI Agents
Built advanced screen parsing tool that converts UI screenshots into structured, actionable elements. Achieved best performance on Windows Agent Arena benchmark using YOLOv8, OCR, and vision models.
VoxIO AI Agent Platform
Architected scalable infrastructure for creating and orchestrating AI agents for enterprise solutions. Built frameworks supporting multi-agent interactions with automated testing pipelines for production deployment.
Browser-Use: AI Browser Automation Framework
Production-grade framework enabling AI agents to control browsers autonomously using LLMs. Supports OpenAI, Anthropic, and Ollama with Playwright for seamless web interaction - clicking, typing, navigating, and data extraction.
LiveKit Voice Agents Framework
Built production-grade realtime voice agent framework with video capabilities. Integrated Deepgram, ElevenLabs, OpenAI, Anthropic, Azure, and Google for STT, TTS, and LLM processing in enterprise conversational AI.
LLM Answer Engine: AI-Powered Search
Perplexity-inspired answer engine providing intelligent search with sources, answers, images, videos, and follow-up questions. Built with Next.js, LangChain, Groq, and Brave Search API for comprehensive RAG-based responses.
FedRAMP Compliance Implementation
Contributed to successful FedRAMP compliance for ExaData Cloud At Customer, achieving 25% security enhancement. Ensured alignment with U.S. government standards and enabled federal contract eligibility.
Open-LLM-VTuber: Voice AI with Live2D
Interactive AI VTuber with voice conversation, Live2D animation, and interruption support. Works completely offline with FunASR, Faster-Whisper, MeloTTS, and MemGPT for long-term memory and animated visual presence.
WORK_EXPERIENCE
Oracle
Senior Member of Technical Staff
- Architected and implemented scalable software infrastructure for VoxIO, a platform specializing in the creation and orchestration of AI agents for enterprise solutions, with emphasis on system interoperability and extensibility
- Researched Gen-AI technologies to develop robust frameworks supporting enterprise-grade AI agent ecosystems, incorporating governance protocols and ensuring compliance throughout the development lifecycle
- Established rigorous quality assurance methodologies for AI agent performance validation, including the design and deployment of automated testing pipelines for complex multi-agent interactions and system integration
- Contributed to successful implementation of FedRAMP compliance for ExaData Cloud At Customer, yielding around 25% enhancement in security measures and ensuring alignment with U.S. government standards
- Developed migration tool that seamlessly transitions on-premise ExaData infrastructure across regions, reducing transition time by more than 50% with guaranteed zero downtime and no data loss
- Collaborated with cross-functional teams to design and implement advanced framework for monitoring Oracle GuestVM operations, resulting in more than 40% reduction in diagnostic time
- Automated data plane events posting process, achieving around 75% reduction in response time for Oracle GuestVM agent server or admin server even during downtime
- Led project to develop SSL/TLS certificate lifecycle management system overseeing CSR, CA, CRT, and CRL handling, alongside managing Keystores and Truststores
- Engineered log monitoring system to gather error logs from distributed systems and send them to central server, enabling SREs to investigate system faults efficiently
- Designed log collection service to collect logs from highly secured customer VMs and upload them to OCI Object Storage bucket via secure PAR URL
Quixote Automotive Technologies
iOS Developer
- Constructed iOS application in MVVM architecture, deployed to App Store, incorporating Firebase, Google Map SDKs for custom navigation with real-time updates and push notifications
- Boosted UITableView and UICollectionView speed by 30% through efficient background threads and lazy loading. Integrated REST APIs and enhanced security with encrypted Core Data