Haroon

Advisor

OudenX • Temps partiel

Feb 2026 - Present • 3 mos

Advisor at OudenX, leading the design and deployment of scalable AI systems, LLM-powered architectures, and cloud infrastructure from concept to production

AI/ML Engineer

GenITeam Solutions • Temps plein

Apr 2025 - Feb 2026 • 10 mos

End-to-End AI Project Leadership: Led a technical team in delivering full-cycle AI solutions across diverse domains, including Healthcare, Gaming, and Legal Tech. Oversaw architecture, development, and deployment of mission-critical models. Game AI & Automation: Automated game testing and simulated intelligent in-game behavior using Unity and ML-Agents. Designed and trained Reinforcement Learning (RL) models (utilizing PPO and Imitation Learning) to control agents in complex 3D environments. LLM & RAG Architectures: Built RAG-style systems and multi-agent pipelines, leveraging deep conceptual understanding of Transformer architectures (Attention QKV, embeddings, tokenization) for advanced inference workflows. ML Operations (MLOps): Managed large-scale ML pipelines for data generation, preprocessing, and model evaluation. Integrated Computer Vision tasks into real-time environments to support game designers and engineers. Backend & Systems: Developed high-performance Python backends using FastAPI and async patterns. Managed Linux environments, containerization, and CI/CD workflows to ensure robust deployment. Cloud Infrastructure: Deployed and scaled AI applications using GCP, AWS, and Amazon Bedrock.

AI Engineer

SecretLabz • Temps plein

Dec 2024 - Mar 2025 • 3 mos

Generative AI Integration: Designed dynamic, context-aware interview simulations that provide personalized feedback using OpenAI’s GPT APIs and advanced prompt engineering. Full-Stack Architecture: Built a responsive end-to-end system integrating a React frontend with a high-performance Flask and PyTorch backend. User Impact: Achieved a 30% improvement in user preparation outcomes, validated through engagement analytics and qualitative feedback loops. Local LLM Deployment: Engineered a desktop voice assistant utilizing the DeepSeek R1 (7B) model via Ollama for privacy-focused, real-time natural language interaction. Multimodal Pipelines: Integrated continuous speech-to-text (SpeechRecognition) and text-to-speech (pyttsx3) modules for fluid, hands-free communication. Performance Optimization: Focused on low-latency inference and prompt optimization to ensure seamless voice-based productivity.