AKSHYA DHIMAN

At ONGC, I architected a dual-model OCR pipeline using Qwen2.5-VL and TATR, achieving 80% precision on handwritten pages.

I also developed a PostgreSQL/pgvector RAG system indexing 10,000+ chunks. It restricts answers to verified context and refuses to answer beyond it.

For LitExplorer AI, I built a local RAG research assistant for scientific PDFs and reduced hallucination to 5.6% on an 18-query benchmark. A hybrid retriever improved Precision@1 from 0.28 to 0.44.

I also built a three-stage Netflix ETL pipeline that prepared 8,800+ records for analysis in Snowflake. I’m an aspiring AI Engineer with hands-on experience in AI systems and backend infrastructure.