AWS Certified - Verify here

Hi, I'm Abubakar

I am an AI Engineer specializing in building and deploying intelligent systems that solve real-world challenges. With a Master’s in Robotics from the University of Maryland, my work is centered on the cutting edge of Agentic AI, Deep Learning, and Computer Vision.

I have a proven track record of designing high-impact models , from custom transformers for 3D object detection [cite: 51] and fine-tuned LLMs for explaining autonomous vehicle behavior , to advanced multimodal systems that fuse LiDAR and image data for enhanced perception. My experience extends to developing multi-agent systems using LangGraph and engineering custom data ingestion pipelines for client-specific knowledge bases.

I am proficient in the entire MLOps lifecycle , leveraging tools like PyTorch, TensorFlow, and a range of AWS services to take complex data and transform it into innovative, production-ready solutions.

Resume

AI Engineer at Easybee AI

Position: AI Engineer

Duration: August 2025 - Present

Key Contributions:

Developing multi-agent systems using LangGraph and CrewAI by assigning specialized roles to agents to automate complex, interconnected tasks and enhance operational efficiency. Deployed the system to AWS ECS for automatic scaling.
Engineered a proprietary custom web crawler and data ingestion pipeline to construct client-specific Knowledge Bases, implementing advanced chunking and data enhancement techniques before indexing into a Pinecone vector database for agentic AI access.
Integrated and managed WhatsApp and Twilio messaging channels using Model Context Protocol (MCP) to enable seamless and continuous communication between customers and the AI system across different platforms.

Deep Learning Engineer at UMD PAL

Position: Deep Learning Engineer

Duration: March 2024 - May 2025

Key Contributions:

Developed a custom vision model based on Masked Autoencoders (MAE) architecture and trained it on RGB orthomosaics from drones to estimate Leaf Area Index (LAI), achieved a never before novel R2 score of 0.87 using only RGB data.
Created further trainable vision models on top of the MAE encoder and outputs to map image features to LAI values, reducing reliance on multispectral hardware and simplified data acquisition and the predicted LAI values are utilized for yield estimation.
Created a ViT model based multispectral pipeline that takes Digital Surface and Terrain Models (DSM & DTM) as inputs and trained it for plant height estimation. The model maps the features to ground truth values, achieving an RMSE of 0.26m, outperforming CNN-based baselines.
Planned and conducted multi-drone missions for collecting high-resolution aerial data across agricultural plots and generated orthomosaics using Pix4D, and created elevation models using Python-Rasterio.
Automated large-scale aerial data preprocessing and normalization pipelines, integrating OpenCV, rasterio, and PyTorch for scalable experimentation.

Computer Vision Intern at Vyorius

Position:Computer Vision Intern

Duration: Aug 2021 - Nov 2021

Key Contributions:

Built YoLO-based video analytics modules for UAVs, enabling real-time object detection, tracking, and classification from aerial streams and enhanced autonomous surveillance coverage by 60% across remote deployments.
Used SOTA segmentation models and integrated them into Vyorius's DataSync Intelligence pipeline to automate aerial image analysis and enhance vision-based landing modules, enabling 40% faster detection of anomalies and regions of interest and contributed to 25% higher precision in GPS-denied landings
Collaborated with cross-functional teams to deploy scalable AI modules into Vyorius's cloud platform, supporting scalable robotic operations across multi-robot missions.