Adit Agarwal
Computer Science
University of Illinois Urbana Champaign
Hi There!
I'm Adit Agarwal, a Computer Science grad from University of Illinois Urbana Champaign. I work at Meta’s Superintelligence Labs, where I helped build the Model Context Protocol (MCP) and agentic orchestration frameworks powering multi-agent systems across Facebook and Instagram. My interests center on intelligent agents, tool orchestration, and scalable AI systems that bridge technology with real-world applications.
Interests
- Human Computer interaction
- Artificial Intelligence
- Reinforcement Learning
- Neural Networks
Education
-
MS in Computer Science, BS in Computer Science & Advertising, Aug 2022 – May 2026
University of Illinois Urbana Champaign, Champaign, IL
Experience
-
Software Engineering Intern – Meta (Superintelligence Labs), New York City, USA
May 2025 – Aug 2025
- Architected and launched Meta’s first Model Context Protocol (MCP)—a standardized context-sharing protocol for LLMs—infrastructure and AI Agent framework, enabling tool orchestration and streamlined data flow.
- Scaled MCP system to support multi-MCP server architecture, integrated 5+ external toolsets, and onboarded three cross-functional engineering teams to leverage the agent for their teams.
- Enhanced a Social Search Agent by designing LLM-based intent parsing & implementing high-performance PHP/C++ retrieval pipelines across Facebook & Instagram—enabling natural language search.
- Developed a new Shopping Agent that leverages MCP to aggregate and rank product listings from Facebook Marketplace and eBay based on user-specified filters, achieving 92% accuracy on agentic evals in matching intent and supporting real-time recommendations.
-
AI/ML Engineering Intern – Droisys (account360.ai), Las Vegas, USA
June 2024 – Aug 2024
- Engineered a custom trained text-to-image latent diffusion model using Low-Rank Adaptation (LoRa) and developed a streamlined UX for designers in the alcoholic beverage industry, enabling generation of advertising creatives up to 90% faster.
- Enabled enhanced data analytics using Image Recognition and sales data, manifesting critical metrics and data-backed suggestions via a user-friendly Gen AI chatbot to help sales reps meet targets.
- Implemented image stitching and object detection with OpenCV & YOLOv8 to enhance shelf monitoring.
-
Software Engineering Intern – Asymtos Co-Distributed Cloud (asymtos.com)
June 2023 – Aug 2023
- Built monitoring tools and interactive dashboards for edge nodes on a distributed cloud system using React, GoLang, and Python. Implemented heartbeat detection, metric collection, plot creation, and more.
- Wrote extensive API documentation detailing various public as well as internal API resources.
Projects
Image Creation Tool
Engineered a custom trained text-to-image latent diffusion model using Low-Rank Adaptation (LoRa) to enhance efficiency. Developed a streamlined UX for designers in the alcoholic beverage industry, enabling generation of advertising creatives up to 90% faster.
Learn More
Data Chatbot
Enabled enhanced data analytics using Image Recognition and sales data. Integrated with YOLOv8 and OpenCV for object detection and image stitching, manifesting critical metrics and data-backed suggestions via a user-friendly Gen AI chatbot to help sales reps meet targets.
Learn More
Generative AI Quizzing Platform
Developed a Generative AI-based quizzing system as a study aid for students using Python, React, LangChain, FAISS, and OpenAI. The platform allows users to upload PDFs and generate customized quiz questions based on user queries, enhancing interactive learning.
Learn More
COVID Vaccine Twitter Bots (India)
Programmed Twitter bots using Python, Twitter API, AWS, and APISetu to autonomously tweet about real-time COVID vaccine availability in 6 major Indian cities. Utilized web scraping techniques to update availability every minute, achieving over 2 million impressions and 2000+ followers.
Learn MoreArticles
Understanding Latent Diffusion Models in AI
A brief summary of the first sample article. This article delves into the fundamentals of latent diffusion models and their applications in modern AI-driven image generation.
Leveraging YOLOv8 for Object Detection in Retail
A brief summary of the second sample article. This article explores the implementation of YOLOv8 for effective object detection and its impact on retail shelf monitoring.
Research
Undergraduate Research Assistant at University of Illinois Urbana Champaign
- Investigating biases in AI models like ChatGPT, BardAI, DallE, and Midjourney under Prof. Hari Sundaram.
- Co-authoring a research paper on smart message framing to encourage reader action.
- Developed web pages, graphics, and analyzed large datasets for research projects.