Driven by passion

to bring

the future of robotics closer.

Portfolio
img

About me

Grad Student @ ASU

Hey, I’m Devika, a robotics and automation enthusiast passionate about blending control theory, machine learning, and innovative engineering to solve real-world challenges! I can help design and optimize robotic systems, automate workflows, and develop solutions that combine intelligence with precision. My work is grounded in analytical problem-solving and a knack for translating theoretical concepts into practical applications. Whether it’s mapping environments for mobile robots, perfecting humanoid gait planning, or streamlining processes with Python scripts, I ensure every task is approached with efficiency and impact. If you're looking for someone who thrives at the intersection of robotics, programming, and creative engineering, I’m here to bring clarity and innovation to your projects.

Download Resume

Professional Experience

Instructional Aide

Arizona State University

(Jan 2025 – May 2025)

Served as an instructional aide for the undergrad course Aircraft Dynamics and Control, which focused on applying control systems to aircraft dynamics. Assisted weekly recitations and led office hours to help students understand core topics such as stability derivatives and linearized equations of motion. Collaborated with the instructor to create goal-aligned instructional materials, while independently revisiting key control theory and dynamics concepts to better support student learning and engagement.

Technical Analyst

Deloitte USI

(Nov 2022 – Jul 2024)

At Deloitte, I designed and deployed end-to-end AWS data pipelines using Python, C++, and Boto3 to automate AMI version report generation across multiple EC2 instances. I integrated Office365 APIs with SharePoint and Microsoft Teams for secure, credential-based data transfer and automated compliance dashboard updates. Additionally, I resolved cross-platform package failures in AWS KPA integrations by identifying dependency conflicts and implementing rollback strategies to ensure system stability.

Robotics Research Intern

AugSense Lab Ltd

(May 2020 – Nov 2020)

As an intern, I developed a high-performance Python pipeline to stitch 2,500 medical images using OpenCV and NumPy, transforming raw slices into seamless composites. I optimized the stitching algorithm with parallel programming techniques to boost efficiency and then deployed the solution on a GPU using PyCUDA for accelerated computation. This upgrade slashed the processing time by nearly 73%, turning a slow task into a rapid and scalable solution.

Recent Projects

...

Generative Modeling with Vision Transformers for Image Super-Resolution

- Implemented and trained three super-resolution models (SwinIR Transformer, Basic ViT, CNN-SRGAN) on the DIV2K dataset using a modular PyTorch pipeline with mixed loss functions (L1 + VGG + SSIM).
- Achieved Peak Signal to Noise Ratio (PSNR) of 25.8 dB and SSIM of 0.75 with SwinIR, outperforming CNN and baseline transformer by ~2.5 dB and demonstrating superior texture reconstruction.
- Enabled video and image inference modes, optimized training with Adam and progressive loss scaling, and benchmarked results quantitatively and visually.

img

De-Fine Tuning : Unlearning Multilingual Language Models

- Fine-tuned LLaMA 3.2B and Qwen 2.5 models on multilingual TOFU datasets (English, Hindi, Korean) to adapt them for downstream translation and reasoning tasks.
- Automated translation pipeline using Mistral API; conducted large-scale fine-tuning and evaluation across multiple unlearning strategies.
- Achieved up to 88% reduction in similarity on forget datasets while maintaining more than 90% retention accuracy, validated across Hindi, Korean, and English datasets.

img
>

Autonomous Maze Solver Using MyCobot Pro 600

- Built an autonomous maze-solving system for MyCobot Pro 600 using computer vision and path planning.
- Created a digital twin in MATLAB and SolidWorks, integrating ArUco markers for accurate coordinate conversion and end-effector positioning.
- Achieved 94% accuracy in maze navigation, demonstrating precision in path formulation and inverse kinematics calculations.

Design and Development of Humanoid Robot

- Designed a 12DoF biped humanoid robot, focusing on lower limbs for human-like walking.
- Developed CAD and URDF models, selecting DYNAMIXEL actuators based on MATLAB simulations.
- Implemented LIPM-based walking gait controller with preview control and interfaced MATLAB with ROS1.

Indoor Package Dispenser Bot

- Co-developed an intelligent differential drive bot for autonomous warehouse movement.
- Simulated a 500 sq. ft. warehouse in Gazebo using LiDAR and G mapping SLAM for navigation.
- Used PRM algorithm to generate 50+ waypoints, optimizing traversal efficiency.

Contact info

Email

devikasnair99@gmail.com

LinkedIn

www.linkedin.com/in/devika-s-nair/