Tariful Islam

At INFINOZ, I architected and developed InfiChat, an AI-driven conversational assistant, owning model integration, system architecture, and end-to-end deployment. I also led R&D on a real-time Voice Agent, bringing speech recognition, natural language understanding, and speech synthesis into one conversational pipeline. I fine-tuned and evaluated Bangla Text-to-Speech models, including dataset preparation and voice-quality evaluation. In my Real-Time ML Inference System project, I engineered an API that achieved sub-100ms inference latency and cut redundant compute by 80%+ with ONNX Runtime and Redis-backed caching. I added observability and deployed it with Docker Compose and Kubernetes. I built an LLM observability platform with automated evaluation for quality, safety, and hallucination detection. I’m seeking to apply my experience in machine learning, NLP, and applied AI research to deliver reliable, production-ready AI solutions.