Ahsen Nazir

• Led production AI systems for document intelligence and LLM automation across 3 integrated systems in an applied research environment.
• Designed and deployed RAG pipelines with embeddings and vector databases enabling semantic search over 10K+ documents, improving retrieval accuracy and reducing latency.
• Built multi-agent systems using n8n orchestrating 20+ step workflows for end-to-end automation, reducing manual effort by 30-50%.
• Developed tool-using autonomous agents integrating 5+ external APIs for real-time decision-making and execution across dynamic workflows.
• Implemented LLM evaluation and API orchestration systems with rate limiting, streaming, and reliability controls achieving 99%+ production stability.

• Deployed deep learning models on custom datasets supporting 1000+ concurrent inference requests under production workloads.
• Built secure, rate-limited AI APIs with authentication and controlled model exposure.
• Developed distributed AI systems using multi-threading and parallel execution for low-latency inference.
• Designed 50+ reusable business logic modules across modular AI services enabling scalable system architecture.
• Delivered end-to-end AI solutions for clients across USA, Japan, Malaysia, with production deployment and monitoring.

• Developed Stable Diffusion-based image animation pipelines using pose and audio inputs across 3+ configurations.
• Optimized ComfyUI workflows (FaceAlignment, MediaPipe, InsightFace), improving generation efficiency and output sta
bility.
• Integrated vision and audio models for multimodal animation systems enabling synchronized generation across 2 modali
ties with real-time inference support.
• Reduced memory usage by 50% via model offloading and compute optimization enabling large diffusion workflows on
constrained GPUs.
• Implemented multi-GPU DiT-style parallelization across 2 GPUs (T4, 1080Ti), improving throughput.