muhammad ishfaq

• Building a multi-agent crowd simulation for an AI-based disaster evacuation platform. • Working across digital-twin environments, agent behavior and reinforcement-learning decision policies. • Turning building and personnel data into personalized, real-time evacuation instructions.

• Built a VRAM-aware inference router using FastAPI and async queues to distribute AI workloads across GPU nodes. • Added checkpoint-based preemption so lower-priority jobs can release GPU memory without losing progress. • Worked with Airflow, Celery, Redis, PostgreSQL and distributed GPU infrastructure.

• Worked with SAP S/4HANA MM/SD across procurement, logistics and inventory processes. • Performed descriptive, predictive and prescriptive analysis on 10 years of urea and DAP sales data. • Delivered data-driven recommendations to the Supply Chain department for demand planning and peak-period preparedness.

• Built air-gapped RAG pipelines and optimized open-source LLMs for constrained, on-premise deployments. • Contributed to computer vision projects involving AI-based image analysis. • Worked on TruckChecker, an AI-based transportation solution, using Python, Docker and open-source AI models.