👋 Hi, I’m @pranaypalem
🧠 AI & Robotics Engineer passionate about building autonomous systems that adapt, learn, and scale
🚀 I specialize in Reinforcement Learning, Robotic Perception, and MLOps pipelines for end-to-end machine learning lifecycle management
🛠 Currently working on deploying scalable deep learning models in real-world robotic and healthcare environments
- Built and deployed deep learning models (Faster R-CNN, YOLOv8, ResNet) for healthcare and robotic vision
- Applied self-supervised techniques like SimCLR and BYOL on medical datasets (NIH Chest X-rays, Mammo)
- Trained RL agents (PPO, TRPO, SAC) in Isaac Lab, MuJoCo, and OpenAI Gym for continuous control and motion planning
- Used Grad-CAM, attention maps, and class re-weighting for explainability and imbalance handling
- Integrated model tuning, augmentation, and real-time inference optimization using PyTorch/TensorFlow + TensorRT
- 🚢 Deployed models in production using Docker, AWS SageMaker, and TensorRT
- 🧪 Tracked experiments with MLflow, managed data & model versioning via DVC and DagsHub
- 🌀 Orchestrated ML pipelines using Apache Airflow + Astronomer
- 🔁 Automated CI/CD with GitHub Actions, ensuring robust model updates and testing
- 📈 Monitored live inference metrics using Grafana + PostgreSQL dashboards
- 🧠 Built ETL + NLP pipelines with HuggingFace and deployed transformer models in real-world workflows
- Developed full-stack robotic systems integrating LiDAR-based SLAM, RRT exploration, and object detection (YOLOv8)
- Simulated dynamic robotic agents using Isaac Sim, Isaac Lab, and MuJoCo (DeepMind)
- Designed quaternion-based foldable robots and improved sim-to-real performance using scikit-optimize
- Integrated navigation and control pipelines using ROS 2, MoveIt 2, Nav2, and URDF
- Deployed Jetson-based inference pipelines with sub-50ms latency
- Developed interactive robotic systems in Unity for perception and control prototyping
Simulated and optimized foldable robot configurations in MuJoCo using adaptive kinematics, servo dynamics, and control tuning for real-world robustness.
Designed an optimization pipeline to identify and tune dynamic parameters of MuJoCo robots for improved sim-to-real accuracy.
Benchmarked PPO and TRPO algorithms using Stable-Baselines3 in OpenAI Gym for continuous robotic control, achieving stable training with 250+ episodic reward.