Ansh User

At Adobe, I proposed and prototyped an on-premise local-LLM alternative to cloud inference for Acrobat AI Assistant. The work addressed privacy and compliance concerns, demonstrated a fully offline, page-cited assistant, and was selected into Adobe's Global Incubator.

I built an automated evaluation pipeline for Adobe Connect Intelligence Q&A using DeepEval and local LLMs, and designed embedding-based query deduplication with Sentence Transformers and FAISS. I also developed CI validation pipelines that reduced manual validation time by 60%.

At HealthKart, I migrated search from keyword retrieval to vector similarity search and implemented RAG with local LLMs to rerank results. I also built query autocorrection and integrated Rasa NLU pipelines for entity extraction.

In my projects, I built a LangGraph video-clipping pipeline with a FastAPI backend and a RAPTOR index summariser using local inference. I'm currently studying for an MSc in Applied Artificial Intelligence at the University of Warwick and seeking AI, machine learning, and software engineering roles in the UK.