The problem Insightz solves

The current work-from-home environment has greatly increased the frequency of virtual meetings, making it urgently necessary to find a more effective way to manage meeting minutes and notes. By using sophisticated speech-to-text conversion and NLP techniques to parse the transcript and extract vital details like the number of attendees, meeting duration, summary of the discussion, and any action items mentioned, this study aims to offer a solution to this problem. Modern NLP and speech recognition algorithms will be used in the suggested solution to accurately record the meeting's details. To make sure the proposed solution satisfies the criteria for an effective meeting management system, it will be assessed using a variety of metrics, including accuracy, precision, recall, and F1 score. To determine the effectiveness of the suggested solution, the results will be compared to those of thecurrently available solutions. A review of the pertinent literature on current theories and research on speech-to-text conversion, NLP, and their applications in various industries will also be included in the study. The potential applications of the suggested solution in the healthcare, educational, and financial sectors will also be covered. The suggested solution will parse the meeting transcript, extract crucial information, and provide a summary of the discussion using advanced speech-totext conversion and NLP techniques.

Challenges we ran into

1. Hardware limitations, since nlp based projects are way too heavy for the system to handel. Gpu is a must but to deploy it using a local server is a difficult task hence time required for the output increases.
2. Limited information or inadequate content offered during meetings gave us a challenge to use better models.
3. Ambiguity and context understanding: NLP models often struggle with understanding ambiguous language or correctly interpreting contextual cues.
4. Bias and fairness: Addressing issues of bias and ensuring fairness in NLP models and algorithms.

We dealt with these problems by brainstorming and trying new algorithms, while the hardware part was resolved by optimizing the code. We tried different models for summarization like bert, bart, t5, etc to deal with ambiguity and fairness of the input to generate an optimal output.