AI-Driven Report Generation through Audio Analysis and Keyword Recognition

This thesis presents the design and development of an AI-based web platform aimed at automating report generation through audio analysis of video recordings. The proposed system allows users to upload videos and associate them with structured reports containing multiple types of keyword fields (basic, multi-choice, tabular, and notes). Through speech recognition and natural language analysis, the audio content is transcribed, segmented, and automatically matched with the relevant fields of the report. The architecture is structured into microservices, including components for speech transcription (based on models such as Whisper, Vosk, and Google Speech API), punctuation modeling, and keyword identification using n-gram and Levenshtein similarity algorithms. Key technologies also include React, FastAPI, RabbitMQ, Firebase, MySQL, PostgreSQL, and Google Cloud Storage. The system significantly reduces manual workload in contexts such as medical visits, audits, or construction site inspections.