Proda-Guide

@Prodago

Nov 2023 - Current

Proda-Guide?

Proda-Guide is a sophisticated chatbot designed to facilitate informed decision-making on the Proda-Guide AI Governance platform. This intuitive conversational interface enables users to seek guidance on governance best practices, compliance, and risk management.

Background

Proda-Guide AI Governance is a cutting-edge SaaS solution that empowers organizations to establish robust governance frameworks for their projects. The platform offers a user-friendly and collaborative environment for risk assessment, operating practice implementation, and progress monitoring.

However, the platform’s complexity and the nuances of governance practices and regulations may lead to user uncertainty. To address this, we leveraged the advancements in AI and chatbot technology to develop a supportive chatbot. The challenge lay in creating a chatbot that could contextualize user queries and provide precise, relevant answers within the intricate domain of governance and compliance.

Experimentation Phase

We initiated the project by conducting thorough research on available chatbot-building tools and technologies, despite having no prior experience in this field. Our exploration included investigating Large Language Models (LLMs) such as GPT-3 and GPT-4, as well as LLM frameworks like Langchain. We leveraged vector database for efficient retrieval and context-aware response generation.

It was fun experimenting with new tools and technologies but the real challenge was to build a production-ready chatbot that can handle the complex queries and provide accurate responses.

Evaluations & Optimizations Phase

Following the experimentation phase, we conducted rigorous evaluations to refine the chatbot’s performance. We assessed various aspects of the RAG pipeline, focusing on context extraction, response generation, and overall accuracy. Addressing the unpredictability of LLM models, which sometimes generate irrelevant responses, presented a significant challenge.

Development Phase

Upon achieving the desired level of performance, we proceeded to deploy the chatbot. We overcame technical challenges by developing a scalable deployment solution and implementing real-time data streaming. Initially, we utilized ChainLit but encountered authentication and data isolation issues. Consequently, we opted to build a custom backend using Django, which is currently in development.