CEO & Co-Founder · TerraByte

Rishi Madhok

I'm building TerraByte - the Earth search engine. Geospatial foundation models that let anyone analyze the live planet in plain language.

Computer-vision researcher turned founder. Previously building satellite & aerial perception at Microsoft, Uber ATG, and Carnegie Mellon's Robotics Institute.

Visit TerraByte Get in touch
Rishi Madhok

About

I'm a computer-vision researcher turned founder. Today I lead TerraByte, where we're building the Earth search engine - geospatial foundation models and a natural-language interface that make satellite imagery as queryable as the web.

Before co-founding TerraByte, I spent nearly six years at Microsoft - rising from applied scientist to Principal Applied Science Manager, where I tech-led AI for the Microsoft Planetary Computer Pro platform and built its first geospatial foundation model and GenAI analytics layer. Earlier I led object-detection, segmentation, and tracking models on aerial and satellite imagery for federal customers. I hold an MS in Computer Vision from Carnegie Mellon's Robotics Institute, and my research spans computer vision and geospatial AI - with work at CVPR, ICLR, NeurIPS, AAAI, and IJCV, plus several granted-and-pending patents.

Experience

Where I've worked

  1. CEO & Co-Founder

    Jan 2026 - Present

    TerraByte

    Building TerraByte, the Earth search engine: geospatial foundation models and a natural-language interface for satellite imagery. Leading product, research, and go-to-market - from Earth-embedding models and similarity search to real-time intelligence products for defense, energy, and finance.

  2. Microsoft

    Feb 2020 - Dec 2025

    5 yrs 11 mos · Seattle, WA · Rose from Data & Applied Scientist to Principal

    1. Principal Applied Science Manager

      Sep 2024 - Dec 2025
      • Tech lead for AI on Microsoft Planetary Computer Pro - set the roadmap, hired and coached a 10-person applied-science pod, and built the platform's first geospatial foundation model plus a GenAI analytics layer.
      • Designed a multimodal Vision-Language + LLM architecture for damage assessment, terrain reasoning, and natural-language tasking.
      • Built a secure RAG stack on GPT-4o / o3 and Azure AI Search, cutting mission-planning query time from hours to minutes.
      • Containerized ONNX / TensorRT models for Jetson Orin edge nodes - 30 FPS full-motion-video inference within a 15W power budget.

    2. Senior Applied Science Manager

      Sep 2022 - Sep 2024

      Led object-detection, semantic-segmentation, and tracking models for federal customers on aerial & satellite imagery, partnering closely with clients to design tailored computer-vision solutions.

    3. Data & Applied Science Lead

      Jul 2022 - Aug 2022

    4. Data & Applied Scientist II

      Sep 2021 - Jun 2022

    5. Data & Applied Scientist

      Feb 2020 - Aug 2021

      Object detection, semantic segmentation, and tracking on aerial imagery.

  3. Graduate Teaching Assistant

    Aug 2019 - Dec 2019

    Carnegie Mellon University · Robotics Institute

    TA for the graduate Computer Vision course (16-720) taught by Prof. John Galeotti - preparing assignments, grading, and holding office hours.

  4. Perception Intern

    May 2018 - Aug 2018

    Uber Advanced Technologies Group · Advisor: Warren Smith

    On Uber's self-driving team, built a model to characterize LiDAR performance and identify the factors driving detector accuracy - informing next-gen sensor specs. Validated transfer-learning approaches so models could be reused across LiDARs with different beam spacings and scanning patterns.

  5. Visiting Research Scholar

    Jun 2018 - Aug 2018

    Carnegie Mellon University · CS Dept · Advisor: Prof. Dave Touretzky

    Built a multi-camera / multi-robot facility for the autonomous robot Cozmo by repurposing old phones as perched cameras. Performed camera calibration and SLAM, and created a server sharing a world map with robot clients for better path planning and navigation.

  6. Graduate Teaching Assistant

    Jan 2018 - May 2018

    Carnegie Mellon University · Robotics Institute

    TA for the graduate Computer Vision course (16-720) taught by Prof. Srinivasa Narasimhan - preparing assignments, grading, and holding office hours.

  7. Research Intern

    Jun 2017 - Aug 2017

    IBM Research Labs, New Delhi · Manager: Dr. Sameep Mehta

    Built a contextual in-video advertising system that places brands in contextually relevant video at the least-intrusive moment - using multi-modal analytics and semantic understanding of video content.

  8. Software Development Intern

    Jun 2016 - Aug 2016

    Shopclues, Gurugram, India

    Owned the notification module of Shopclues's POS app, integrating Firebase Cloud Messaging and Notifications, plus Firebase Analytics for event logging.

Selected Work

Things I've built

TerraByte · Flagship

The Earth Search Engine

Geospatial foundation models that turn raw satellite imagery into searchable embeddings - find oil tanks, deforestation, power plants, or copper mines anywhere on Earth with a plain-language query. Real-time ingestion, predictive AI, and developer APIs.

Foundation modelsEarth embeddingsSimilarity search
terrabyte.ai
Intelligence · Demo

Strait of Hormuz Maritime Intelligence

A multi-intelligence platform for the Strait of Hormuz: Sentinel-1 SAR vessel detection fused with Brent crude pricing and news/OSINT, topped with an automated Opus 4.8 analyst brief that reads the signals and writes the assessment.

Sentinel-1 SARVessel detectionOSINT fusionLLM analysis
Live demo GitHub

Research projects

Sensor Fusion with Single-Photon Detectors

Advisor: Prof. Matthew O'Toole · CMU

A suite of sensor-fusion techniques built around single-photon avalanche diodes (SPADs). Developed a novel algorithm to fuse single-photon LiDAR, stereo cameras, and radar into an intermediate cost-volume representation that, passed through a deep network (PSM-Net), yields better disparity and depth estimates.

Action Recognition using Synthetic Data

Advisor: Prof. Kris Kitani · CMU

Recognizing multi-object activities - like cars turning or making U-turns - by generating synthetic data in Unreal Engine to match real-world distributions, then deploying a bi-directional RNN to classify the activities.

Semantic Understanding of Video

Advisor: Dr. Rajni Jindal · DTU

Preserved both context and temporal sequence of video through a sequence-to-sequence model - an LSTM acting as both encoder and decoder - to generate natural-language summaries of video content.

Publications & Patents

Research

295Citations
7h-index
6i10-index
Google Scholar →

Recent · 2024-2026

Patents

Earlier work

View the full list on Google Scholar →

Skills

What I work with

AI & Machine Learning

  • Foundation models
  • LLMs
  • Computer vision
  • Deep learning
  • Self-supervised learning
  • PyTorch

Geospatial

  • Satellite & SAR imagery
  • Remote sensing
  • Earth embeddings
  • GIS

Engineering

  • Python
  • C++
  • Distributed training
  • APIs & MLOps
  • Cloud (Azure / AWS)

Leadership

  • Team building
  • Product strategy
  • Research direction

Education

Where I studied

Carnegie Mellon University

MS, Computer Vision · Robotics Institute

Aug 2018 - Dec 2019 · CGPA 4.17 / 4.33

Delhi Technological University

BTech, Computer Science & Engineering

Aug 2014 - May 2018 · CGPA 9.368 / 10

Contact

Let's build something that sees the planet.

Whether it's TerraByte, geospatial AI, or computer vision - I'd love to hear from you.

LinkedIn