Aditya Sahay

I've built production-grade AI systems spanning RAG pipelines, fine-tuned transformer models, and autonomous perception systems.

For a transformer-based news summarization project, I fine-tuned facebook/bart-base on 300K CNN/DailyMail samples, deployed a FastAPI REST API, and measured decoding speed-versus-quality tradeoffs. I also built hallucination detection through named-entity overlap and identified cross-article confusion as a primary failure mode.

I built an end-to-end RAG system over 50 arXiv papers using PyMuPDF, FAISS, semantic chunking, and cross-encoder re-ranking. It reduced hallucination by 57% and improved ROUGE-L by 81% over a no-retrieval baseline.

As a Web Development Intern at PuranChand Foundation, I built the foundation's complete Django website independently, including the database schema, backend logic, admin interface, and Render deployment. I'm a final-year B.Tech CSE (AI/ML) student seeking AI/ML engineering work on real problems at scale.