I'm an AI Engineer at Otsuka Corporation in Tokyo, where I work on large language models, agentic systems, and retrieval-augmented generation for enterprise use cases. Most of my day to day is spent fine-tuning models with SFT and RLHF, deploying vLLM-served inference on internal GPU clusters, and building RAG pipelines that hold up under real production traffic rather than just demo conditions.

A lot of my recent work has centered on knowledge systems: agentic RAG chatbots serving over a thousand monthly users, an internal knowledge management platform that auto-generates QA pairs from hundreds of documents, and evaluation pipelines that make it possible to actually measure whether a retrieval setup is getting better or just getting different. I care about the unglamorous parts of AI engineering as much as the modeling itself: latency, access control, evaluation rigor, and the plumbing that makes a model useful to people who aren't machine learning engineers.

Outside of work, I build and open source a few things I wish existed when I needed them. Agent Forge is a multi-agent chat platform with persistent memory and tool calling. The RAG Benchmark Toolkit is a YAML-driven harness for comparing chunking strategies, retrievers, and rerankers across a hundred plus configurations instead of guessing. I also maintain a dataset augmentation pipeline for generating SFT and preference data at scale. I studied Computer Science at IIT Patna, and my research on transformer models for disaster event extraction was published in Online Social Networks and Media.

Before all of this I spent a lot of hours on competitive programming, which is probably why I still enjoy debugging something down to the exact line rather than accepting "it works on my machine." I'm always happy to talk about LLM fine-tuning, RAG architecture, or agent design, so feel free to reach out.

You can find my work at rahilop.github.io/profile, reach me at rahilrizvi0786110@gmail.com, or connect on LinkedIn.