Denim Patel
Specializing in computer vision, robotics, and autonomous systems, building intelligent systems that see, learn, and adapt.
Full ML pipeline, from research to production
I'm a Senior Machine Learning Engineer at SimpliSafe, specializing in computer vision, robotics, and autonomous systems. My passion lies in developing intelligent systems that can perceive, understand, and interact with the world.
My experience spans the full ML pipeline, from research and prototyping to production deployment, with a focus on object detection, tracking, visual odometry, and SLAM. Earlier work at iRobot and Samsung Research America took me from ground-truth localization systems to 3D object detection presented at CES.
- Based in
- Boston, MA, USA
- Currently
- Senior ML Engineer, SimpliSafe
- Focus areas
- Computer Vision · Robotics · SLAM
- Education
- MS Robotics Engineering, WPI
- denimpatel2020@gmail.com
Where I've built things
Six years across three teams, moving from applied research to systems running on real robots and real homes.
- Led the creation of a comprehensive ground truth system for evaluating and quantifying mobile robot localization performance.
- Facilitated seamless information exchange among mobile robots present in the same environment.
- Implemented successful 3D object detection and classification techniques customized for kitchen settings.
- Work presented at CES 2021.
Academic foundation
Master's Degree, Robotics Engineering
B.E., Mechanical Engineering, Minor in Design Engineering
Tools of the trade
Computer Vision & ML
Frameworks & Languages
Robotics
Platforms
Selected work
AI agents, developer tools, robotics, and computer vision projects spanning perception, planning, and hardware, most with code linked below.
micro-harness
An AI agent harness distilled to its essence in a single file. Read it in one sitting and understand what makes an agent an agent.
macro-harness
A production-shaped AI agent, built layer by layer: streaming, retries, memory compaction, permissions, sandboxing, and tool integrations, each one explained.
Zion
Turn any repository into an explorable 3D city. Folders are districts, files are buildings, and hotspots, ownership, and missing docs show up in the skyline. Pure Python standard library, no install step.
Laya: Command-Safety Benchmark
A new use case for the open-source Laya decision model: gating which shell commands a coding agent may run. I built a 1,237-command labeled benchmark from GTFOBins, LOLBAS, and nl2bash, measured 0.79 AUC zero-shot, and analyzed where it fails.
Visual Odometry
A mono-camera visual odometry pipeline covering feature detection, feature tracking, and pose estimation.
Object Detection: Transparent Globe
Detection and localization of a transparent globe, comparing classical feature-based techniques against a Faster R-CNN baseline.
Kalman Filter Object Tracking
Object tracking using a Kalman filter with a constant-velocity motion model and a shape-detection-based measurement model.
Camera Calibration
Gold-standard camera calibration methods, applied to place an AR object on a chessboard using the calibrated camera.
Motion Planning (ANA*)
Dijkstra, A*, and anytime A* variants implemented from scratch. ANA* refines path quality with each iteration.
SCARA Robot
Designed and developed a SCARA robot for a custom application, integrating 3D printing with frameworks like Marlin and Repetier.
OpenCV: Face Recognition
Applied OpenCV's built-in algorithms, including Haar cascades, to build a facial recognition pipeline.
Rapidly-exploring Random Trees (RRT)
RRT and its variants (RRT* and Informed RRT*) for path planning in higher dimensions, refining path quality each iteration.
ROS MoveIt!
Optimized the pick-and-place operation of a robotic arm based on the physical features of target objects.
Mechatronic Product Development
Developed a pan-tilt motion module optimized for low inertia and rigidity, with all motors mounted on the base.
Age Prediction from Image
Predicting a person's age from a single image using a trained deep learning model.
Published GitHub Pages apps
Standalone interactive web apps I've designed, built, and shipped end-to-end, each hosted on GitHub Pages.
GitLearn
Curriculum-based, interactive way to learn Git: guided lessons with a live visualization of your repo state, plus a free-form Playground mode.
Interactive Courses
255 interactive guides spanning AI, vision & geometry, and math. Build real intuition by seeing every concept animated, not just reading about it.
diff-learning
Build a GPT from an empty file, one diff at a time. Step through each lesson's code change, then run and tweak every step right in your browser.
Transformer Inference Roofline Explorer
Interactive, learning-first visualizer for Transformer inference roofline analysis: arithmetic intensity, batch size, latency, and cost tradeoffs.
Interactive Information Theory
Build intuition for probability, information content, entropy, KL divergence, and channel capacity using a biased coin as a running example.
Black-Scholes Sensitivity Analysis
Interactive exploration of Black-Scholes option pricing sensitivities (the Greeks) as volatility, time, and market parameters change.
EisenMatrix Calendar
A calendar that helps you choose what to work on next, using Eisenhower Matrix prioritization and drag-and-drop tasks, integrated into a full calendar view.
AI
Informational site charting the history of AI: key milestones from the Turing Test to AlphaFold, plus notes on robotics, ROS, and computer vision.
Image Editor
Browser-based image editor with filters, adjustments, and cropping tools. No server round-trip required.
Research Paper Feed
A curated, feed-style browser for discovering and tracking new research papers.
Awards
Let's talk about what you're building.
Open to conversations on computer vision, robotics, and ML systems roles, or just a good technical chat.