Belal – AI agent development, Python, LLM

A project focused on building a self-serve code modernization platform for migrating and upgrading full codebases across languages and frameworks using coordinated multi-agent workflows. The platform combines MCP-powered repository navigation, code editing, and automated build, run, and test execution with a repository-aware RAG system that retrieves architecture-level context—such as entry points, dependencies, and module boundaries—to improve correctness and overall modernization quality.

• Architected backend and frontend systems, establishing a scalable foundation for a self-serve code modernization platform.
• Engineered a repository-aware RAG system customized to client codebases.
• Designed and implemented an orchestrator agent coordinating a pool of specialized sub-agents.
• Built webhook-driven listeners to automate pull request creation.
• Developed an end-to-end evaluation pipeline to quantify code quality and modernization outcomes.
• Implemented frontend features and reusable UI components to accelerate product expansion.

A project focused on leading the development of AI-powered customer automation solutions for automotive retail. The solution combined LLM agents, tool-integrated workflows, and API-first services to improve lead conversion, reduce churn, and lower inference costs by 85%–95% in production.

• Built and deployed LLM-powered customer automation systems, improving lead conversion and reducing churn in production.
• Reduced chat processing costs by 85–95% through prompt and tool routing optimization.
• Developed agentic workflows with LangChain and GPT/Llama-class models for tool-integrated automation.
• Delivered API-first backend services using FastAPI and Flask for frontend and third-party integrations.
• Deployed scalable production infrastructure on AWS with Docker, Kubernetes, Nginx, Redis, and ClickHouse.
• Established CI/CD pipelines to enable faster delivery and more reliable releases.

A project focused on developing end-to-end computer vision and machine learning platforms for classification, segmentation, and recommendation workflows. The solution leveraged PyTorch and TensorFlow, with scalable pipelines for training, inference, experimentation, and evaluation, and was deployed to support millions of real-time predictions per day.

• Led development of computer vision systems for classification, segmentation, and recommendation.
• Built end-to-end training and inference pipelines with PyTorch and TensorFlow.
• Delivered production services supporting millions of real-time predictions daily.
• Improved model quality through data pipelines, augmentation, and offline/online evaluation.
• Built scalable experimentation workflows on AWS.

A project focused on leading the development of AI-driven inspection pipelines for critical infrastructure. The solution delivered scalable computer vision and geospatial analytics systems that identified and mapped structural defects on dams and bridges using high-resolution aerial imagery.

• Led CV pipelines for detecting and mapping defects on dams and bridges using drone imagery.
• Developed semantic segmentation and defect classification models.
• Built end-to-end geospatial analytics workflows for infrastructure inspection.
• Developed scalable processing pipelines with GDAL, rasterio, and QGIS.
• Ensured CRS and projection correctness across geospatial workflows.

A project focused on developing end-to-end remote sensing machine learning pipelines for object segmentation in multispectral satellite imagery. The solution combined large-scale raster processing, georeferencing, projection transformations, and DEM generation to support high-resolution geospatial workflows.

• Built remote sensing ML pipelines for object segmentation in multispectral satellite imagery.
• Implemented large-scale raster processing and georeferencing workflows using GDAL and Rasterio.
• Developed projection transform and DEM generation workflows for accurate geospatial analysis.
• Optimized training and inference pipelines for high-resolution geospatial datasets under strict latency and throughput constraints.
• Supported scalable processing of satellite imagery for production remote sensing workflows.