At Drytis, I provide live and async engineering support to US clients, including pair programming, debugging, refactoring, and feature completion. I use LLM-assisted development and prompt engineering, and document sessions and code changes to security standards.
At Loop Techno Systems, I designed a four-tier FAISS IVF-PQ retrieval platform over 100K+ embeddings, reaching 86% top-5 recall at under 50ms. Its document Q&A supports PDF, DOCX, XLSX, and CSV files, with reranking for cited answers.
I also built a three-stage LLM output validator that cut hallucinations from 22% to 8% across 10K production queries. My work there included raising ROUGE-L from 0.31 to 0.43 and shipping a video-diffusion toolkit for coherent HD video up to 4K.
As a Predoc Research Fellow at IIT Roorkee, I built an automated thermal-landmark pipeline for the SF-TL54 research paper, achieving mean IoU of 0.8012 against a 0.5446 fixed-ROI baseline. I also built a Raspberry Pi multi-sensor logger with synchronized capture and a SQL schema for ML.
