Keya Gohil

Built a scalable PySpark pipeline to preprocess 1,350+ motor vehicle fatality records, labeling incidents in the top 10 accidentprone ZIP codes.

Developed an AI-powered research assistant leveraging Python, LangChain, and LLM APIs to automate web research, aggregate information, and generate structured reports with dynamic multi-source data retrieval.

Implemented the ROME method to apply targeted edits to a large language model (gpt2-xl). The experiment focused on inserting a counterfactual fact—changing the model’s internal knowledge.

Developed a Knowledge Graph-driven Retrieval-Augmented Generation (RAG) system using Neo4j, Llama 3.2, and Gemini-Flash 2.5, enabling context-aware and conversational answers over a structured movie dataset.

Engineered a Netflix-like platform using a microservices architecture with separate servers for India, USA, and UK movies, integrated with Kong API Gateway for Routing, Prometheus for monitoring, and robust security measures.

Optimized Hyperparameter of the Graph Network-based Simulators (GNS) to achieve 95% accurate prediction in substantially less time and computational resources.

Led a team to develop web-based applications for processing diverse documents types. The chatbot is nearly accurate in answering logical questions and factual questions.

Trained 3d segmentation models like nnUnet and monai SegResNet and achieved an average pseudo dice of 0.85 and 0.75 respectively for different affected regions of the Brain Tumour.

Engineered a Python-based web app that enhances text input and achieves more than 90% accurate auto-correction and predictive text generation.

Developed a web-application for sharing documents which provides 100% reduction in unauthorized data access in a local server system in a fully automated manner.