Deep Mathukiya

Built a LangGraph/DSPy multi-agent framework with human-in-the-loop code generation and validation; designed a modular system for registering and integrating financial data sources.

Created a RAG platform using NVIDIA Nemotron-3 and Milvus for semantic PDF search; integrated vector retrieval with an AI chatbot for accurate document question answering.

Achieved 88% flood risk prediction accuracy by combining rainfall time-series forecasting with elevation and drainage data; crafted GIS maps with image segmentation for more than 100 communities and an early warning web platform.

Developed a solution to translate Gujarati Sign Language into speech and vice versa. Curated a custom dataset for Gujarati Sign Language, as no public datasets were available. Used AR technology for speech-to-sign generation, converting spoken words into visual signs. Achieved real-time translation with a CNN model for gesture recognition and AR for sign visualization.

An AI-driven platform designed to assist farmers in identifying crop diseases using a CNN model for leaf analysis, with an integrated AR feature for scanning crops and offering detailed disease insights and solutions.

Flood Prediction Application: A web app providing real-time flood risk analysis through flood mapping, machine learning-based flood-prone region predictions, and user alerts for early warnings, enabling proactive decision-making.It also Takes Picture from satelite and highlite the flooded area.

This Java-based Android application enables users to choose a .txt file from their device. Upon selecting a text file, the application utilizes the Python gTTS library to convert the text content into an .mp3 file.