Abilmansur Ashim

I specialize in building end-to-end solutions: from data collection and model training/fine-tuning to production deployment with monitoring and CI/CD. In recent projects, I improved retriever Recall@10 from 0.68 to 0.86 by fine-tuning a bi-encoder on domain-specific data, reduced RAG pipeline latency from 3.8s to 1.7s, and automated approximately 35% of incoming requests using an agentic assistant with hybrid RAG and tool calling. My stack: Python, PyTorch, Transformers, LangChain/LangGraph, FastAPI, Qdrant, Docker, Kubernetes, MLflow, Prometheus/Grafana.