richag@mit.edu | +1 6177809005
SELECTED PROJECTS
My research focuses on the emergent behaviors of generative AI, particularly how it often overlooks humans as active users. I aim to bridge this gap by pairing generative AI outputs with actionable insights, shifting AI from a visualization tool to a data-backed decision-making system. (available upon request)
Curator AI - Speech, AI & AR for E-Commerce
(MIT Generative AI Hackathon Winner)
Created a voice-controlled AI recommendation system with visual contextual understanding for Furniture
Market, integrating Whisper API, ChatGPT-4o, Google APIs, RoomPlan API, and AR development tools.
Revolutionized user experience by enabling seamless interaction and enhanced product discovery.
ACADIA 2023 - Workshop Tutor
(Hybrid Making: Physical
Explorations with Computational Matter)
Text-to-Robotic Assembly for Functional Objects
(Best Poster Award - 2.156 AI/ML for Eng. Design)
Developed a pipeline combining 3D generative AI and Vision-Language Models to automate component
assignment for robotic assembly.
Achieved 98% accuracy in identifying functional components, enhancing user selection precision.
3DGeometryTool - 3DIT
A React Three Fiber application utilizing Zustand for state management and Leva for UI controls to dynamically manipulate 3D cubes and their vertices with intuitive TransformControls.
Responsible AI
(MIT AI Decision-making and Society)
Bias detection, red teaming and evaluations in modern Artifical Intelligence system.
Sketch to 3D software
(1st Prize, HMC Architects’ Hackathon, LA 2021)
Led “form sketcher” workflow (Part 1/3), analyzing environment alongside M. Mistry & T. Bhattacharjee
Aimed for flexible any age professional design integrating technology seamlessly with creative processes
Built a Glitch-based websocket to bridge p5.js data (sketch/voice/vision) with open-source ML in
Rhino-Hops-Grasshopper
ArchiDAO
(Co-founder)
Peer-to-peer Network on Blockchain for Architects, Designers and Technologists.
Neural Network Project for Image Completion
Fine-tuned a specialized Stable Diffusion model on 10k meticulously curated image pairs.
Incorporated advanced CLIP-based guidance and custom losses for robust 0.95811 alignment.
Explored adversarial noise/occlusion vulnerabilities, refining pipelines for enhanced resilience.
Arch-SaaS
(Product Development - ideation to launch)
Product for single source of truth - Chaos to Efficiency
PLUS AI
(MIT Media lab- Gold team, MIT Sandbox)
AI Ecosystem for Architecture and design.
Plus +’ goal is not just to make our work more efficient, but sharing our ways of thinking and designing with the world, all while keeping our agency and intent.
Flex the muscle, solve the puzzle
(MIT Web Lab)
Built a gesture-controlled AI puzzle with React.js and Gesture recognition via Google’s teachable machine.
Realtime Mixed reality and IOT for Cooking Assistance