Zohair – LLM, Python, AI

A secure AI platform enabling access to 50+ LLMs with a modular architecture for easy model integration. Developed a React-based UI where users could create projects, upload files, and interact via chat with vectorized content indexed in OpenSearch.

• Created a novel architecture to simplify AI and RAG usage, enabling users to fully utilize the system without technical knowledge;
• Designed the architecture in a modular fashion, allowing independent Lambda function sets to be easily swapped with improved frameworks (e.g., replaced LangChain with LlamaIndex or changed the vector database) without altering the core infrastructure;
• Implemented scalable solutions such as LLM load balancing to support thousands of concurrent users across multiple LLM models, overcoming typical TPM/RPM limitations;
• Structured the frontend with modular components, enabling seamless redesign and refactoring without disrupting user experience.

A modular agentic framework extending Microsoft’s Autogen 0.6 to support highly flexible and configurable AI workflows. The product enabled users to design complex multi-agent systems using JSON, a drag-and-drop UI built with React Flow, or even natural language input. Agents communicated via a pub/sub model, allowing for parallel execution, while a system of hierarchical “managers” coordinated agent behavior based on task relevance. The solution supported deep nesting for advanced use cases and was deployed using AWS Lambda (Python) with a React-based interface that could be embedded into existing platforms.

• Designed and developed a highly flexible agentic system that extended the potential of an open-source framework, making it accessible for non-technical users to build custom agent workflows;
• Enabled support for any LLM backend, with added capabilities for retrieval-augmented generation (RAG) and multimodal inputs;
• Built a user-friendly visual interface using React Flow, allowing users to intuitively map out agent interactions and data flow;
• Ensured horizontal scalability and cost-efficiency by deploying agents as AWS Lambda functions, minimizing infrastructure overhead;
• Streamlined the system architecture to support scalable, modular, and customizable agent-based flows for a variety of advanced use cases.

A personalized push notification system for an educational institution, enabling targeted messages to students based on various attributes and real-time events. Built with a React frontend and a serverless Node.js backend using AWS Lambda, OpenSearch, and DynamoDB. It integrated with enterprise data systems to deliver context-aware alerts—for example, notifying students about academic performance changes or guiding them to seats via geolocation. The system was embedded in the student mobile app and used campus-wide IoT devices for real-time, location-based triggers.

• Integrated deeply with enterprise platforms such as Salesforce, Oracle PeopleSoft, and Mulesoft, enabling a unified interface to select user segments and preview message recipients in real time;
• Built robust analytics to track message delivery, engagement, and interaction patterns, providing transparency and actionable insights;
• Significantly reduced communication turnaround time from days to minutes, while scaling automated notifications to ensure students received timely, relevant information;
• Introduced novel integrations with IoT devices like BLE beacons, merging hardware and software to trigger location-based messaging across campus environments.

A comprehensive mobile application designed to serve as a central hub for students throughout their academic journey, providing seamless access to assignments, grades, campus resources, events, maps, and interactive features. Built with React Native and deployed on a modern, serverless AWS architecture, the app was engineered for performance and scalability—reliably supporting over 100,000 concurrent users. It also integrated with native device wallets to enable secure digital access to facilities, offering a smooth and connected campus experience.

• Architected and developed the entire frontend and backend of a highly acclaimed university mobile app, recognized as one of the best at the time;
• Led core development efforts and mentored junior developers on specific feature implementations;
• Ensured the app was resilient and maintained excellent performance, including on older devices;
• Contributed to the app’s user-friendly design, resulting in high daily active user engagement;
• Supported multiple offers from ventures interested in buying or licensing the app, reflecting its market value and success.