🖥️MS-CSE Student at UC San Diego📊Ex Data Scientist - III at Walmart🎓IIT Madras 2021 Alumni

Hi, I'm Ayush Maniar. Currently, I'm pursuing a Master of Science in Computer Science (MSCS) at UC San Diego, with a specialization in Artificial Intelligence. I am passionate about AI Agents, particularly in the context of Multi-agent systems and Agentic-RAG. My research focuses on multi-agent collaboration and reasoning, with work accepted at NeurIPS 2025. I'm also the creator of an open-source PowerPoint MCP server that enables AI assistants to automate presentation creation with LaTeX rendering and real-time editing capabilities.
Recently, I completed a Research Scientist Intern - Generative AI position at Riot Games, where I developed AI creative story writing engines using dual-agent architectures, built story-to-code pipelines for game quest systems, and created multimodal AI comic book generators. Before joining UCSD, I worked as a Data Scientist III at Walmart Global Tech India (SAMS Club), where I led multiple high-impact projects creating over $120M in business value. My work included developing patented time series algorithms for transit time prediction, delivery delay prediction models, and optimization solutions using Mixed Integer Linear Programming.
During my undergraduate studies at IIT Madras, I co-founded 'Eye In The Sky', a Deep Learning & Computer Vision startup focused on disaster management using drones. The startup raised INR 19.5 Lakhs in funding, won the Microsoft Codefundo++ 2019 challenge (among 2000+ teams), and received the International Microsoft 'AI For Earth' Grant. My other interests include Gaming, Augmented Reality, and Bioinformatics.





Showcasing research and development work in AI Agents.
This research project, conducted as part of my MS Research under Prof. Lianhui Qin at UC San Diego, explores multi-agent collaboration in complex environments like Minecraft. The work has been accepted at NeurIPS 2025.
The framework enables multiple AI agents to reason, plan, and work together through peer-peer collaboration tasks that test different aspects of collaborative reasoning such as sharing resources, skill sets, expertise, and information. We leveraged the efficient function calling abilities of LLMs to ground the agents in the world of Minecraft by creating custom tools, and conducted comparative evaluations of contemporary language models, highlighting success rates across tasks and providing insights into multi-agent collaboration efficiencies.
A powerful Model Context Protocol (MCP) server for PowerPoint automation that enables AI assistants like Claude to create, read, and edit presentations in real-time. Features template-first design, bidirectional COM automation, LaTeX equation rendering, HTML-style formatting, and controllable animations. Unlike write-only libraries, this provides true real-time editing capabilities with multimodal slide analysis.
claude mcp add powerpoint -- uvx powerpoint-mcpResearches topics and builds decks with citations automatically.
Analyzes datasets and creates free-form Python plots directly on slides.
Reads local files to understand structure and generates architecture presentations.
Populates corporate templates with data from CSVs or other sources.
Creates lecture slides with complex mathematical equations rendered via LaTeX.
Turns presentations into interactive tutors that explain slides and quiz users.
I'm working on adding more of my projects here. Check back soon!