"Unlocking the Future: The Advanced AI Chatbot"

Diving deeper into AI, I often ponder the reasons behind businesses' growing interest in this technology. Each AI project is born with a purpose, and my current pro bono initiative is no exception. It's designed to empower a small organisation to leverage their own data and information effectively, turning it into actionable insights and answers to critical questions. The primary exposure many of these organisations have to AI comes through tools like ChatGPT, and they see this as a promising opportunity. I'm here to offer strategies that will not only address current needs but also help them plan for the future. I'll share insights into this approach below.

Project Update: Enhancing Organisational Capabilities

My journey began with developing a ChatGPT-like chatbot using OpenAI technologies. Now, I'm evolving this project to create a more advanced version. This new iteration aims to broaden the variety of document types the chatbot can analyse, thereby enhancing the accuracy and reliability of its responses. The world of AI has progressed, presenting us with newer and more efficient question-and-answer models that aren't limited to OpenAI's offerings. Embracing this diversity is crucial for not being tethered to a single model, enabling the organisation to enrich their data repository with a mix of structured and unstructured texts, images, and videos. This comprehensive data library, along with the chatbot's ability to conduct internet searches, ensures well-informed responses to user inquiries.

For the Developers: Behind the Scenes

The chatbot is being crafted using a variety of technologies. Here's a glimpse into the technical landscape:

• The foundation is set within a Conda environment, using Python for programming.
• While OpenAI models serve as the initial framework, the system is designed for flexibility, allowing integration with any large language model as needed.
• The Milvus Vector Database is the chosen database, with openness to other vector databases.
• Supporting technologies include Llama Index and Langchain, enhancing data retrieval and module connectivity.

Moreover, I'm utilising OpenAI's chat completion APIs, focusing on maintaining conversation history for contextual relevance. Llama Index helps bridge the chatbot with the Milvus Vector database for improved query processing. To digitise ancient manuscripts, pytesseract comes into play, converting image-based documents to text. Furthermore, I'm leveraging moviepy, speech_recognition, and googletrans to transcribe and translate video recordings into text, with Langchain knitting the various modules into a cohesive unit. The ultimate goal is to render every content piece searchable by the Question and Answer chatbot, unlocking new dimensions of accessibility and insight for the organisation.

It’s an interesting project which I enjoy completing in my spare time while working on a digital transformation programme at Westminster Abbey. If you have any suggestions or cutting edge insights about improving the objective above, please let me know.