Computer Vision • Deep Learning • Robotics Software
Building reliable intelligent machines from first principles—from model fundamentals to deployable systems.
I am a systems-oriented AI/ML builder focused on the path from visual perception and deep-learning fundamentals to dependable robotics software. I care about understanding how models work, measuring their behavior, and integrating them into systems that can operate under real-world constraints.
My work currently spans machine-learning foundations, governed AI/RAG systems, Python and C++ engineering, MLOps, ROS-based robotics, sensors, and real-time data workflows. The next stage of this portfolio is centered on camera-based perception for autonomous machines.
Principles: fundamentals first · measurable experiments · clear documentation · reliable systems
| Area | What I am building toward |
|---|---|
| Computer vision | Image pipelines, object detection, visual tracking, camera calibration, and perception evaluation |
| Deep learning | From-scratch understanding, reproducible experiments, representation learning, and model deployment |
| Robotics | ROS 2 software, autonomous mobile robots, sensor integration, simulation, and real-time behavior |
| Systems engineering | Python/C++, APIs, testing, observability, CI/CD, and MLOps for production-minded ML |
- AMR_template — ROS-oriented autonomous mobile robot project template with CMake, package configuration, launch files, and simulation-world structure.
- Robotics projects with documentation — Hands-on robotics learning across Arduino, Python, sensors, real-time visualization, and engineering concepts.
- Linear Regression in Python and C++ — A from-scratch implementation in two languages, emphasizing mathematical foundations and low-level control.
- maitri_model — Privacy-conscious AI/RAG application with a FastAPI backend, Next.js frontend, governed retrieval, local inference, evaluation gates, and telemetry.
- RAG — Retrieval-augmented generation experiments covering embeddings, vector databases, memory, and model orchestration.
- data-cleaning-pipeline — Reusable data preparation components for loading, normalization, missing values, duplicates, and outlier handling.
I am consolidating these foundations into a robotics perception portfolio: camera input → preprocessing → deep model inference → tracking and state estimation → robot decision-making. The goal is not only to train a model, but to make the full system reproducible, testable, observable, and useful on a robot.
Understand the mathematics
↓
Build a minimal implementation
↓
Measure with reproducible experiments
↓
Integrate into a tested system
↓
Document the trade-offs
The best way to follow my work is through the repositories above. I am especially interested in computer vision, deep learning for embodied systems, ROS 2, sensor fusion, and the engineering required to move research ideas into dependable robotics software.