Project Introduction
HLD
Discovery Phase Details and Process
Stakeholder Meetings
Conducted initial meetings with key stakeholders, including consultants, technical leads, and AI specialists, to gather requirements and understand the main point.
Current System Analysis
Evaluated existing systems and processes to identify integration points and potential challenges.
User Persona Development:
Created detailed user personas to understand the end-users better, including scholars, educators, and learners.
Workflow Mapping:
Mapped out current workflows to identify areas for improvement and integration points for the AI solutions.
Pilot Planning
Developed a detailed plan for a pilot phase, including objectives, success criteria, and timelines.
Requirement Gathering
Functional Requirements: Documented specific functionalities needed, such as speech-to-text conversion, accurate translation, voice cloning, and lip-syncing.
Non-Functional Requirements: Captured performance expectations, scalability needs, and security considerations.
Technology Evaluation:
AI Model Assessment: Evaluated various AI models for summarization and content recommendation capabilities.
UI/UX Considerations: Reviewed best practices for UI/UX design to ensure a user-friendly interface.
Feasibility Study: Conducted a feasibility study to assess the technical, operational, and financial viability of the project.
Libraries Used
TensorFlow and PyTorch: For fine-tuning the Gen AI models, both TensorFlow and PyTorch were used due to their flexibility and robust community support.
Transformers (by Hugging Face): Essential for leveraging pre-trained models and implementing the Retrieve and Generate (RAG) system.
Scikit-learn: Used for preprocessing call transcripts and implementing various machine learning tasks during the fine-tuning process.
Flask: Utilized for setting up the backend services, providing a lightweight and scalable solution for API development.
React: Chosen for the frontend development of the Copilot UI due to its component-based architecture and efficiency in building interactive UIs.
Databases Used
MongoDB: Used as the primary database for storing call transcripts, user data, and interaction logs. Its flexibility and scalability made it an ideal choice for handling large volumes of unstructured data.
Elasticsearch: Implemented for the RAG system to efficiently index and retrieve knowledge-based content, ensuring quick and relevant responses.
PostgreSQL: Utilized for managing structured data, including user profiles, system configurations, and administrative settings. Its reliability and advanced querying capabilities were crucial for maintaining robust data integrity.
Redis: Employed for caching frequently accessed data and managing session information, enhancing the overall performance and responsiveness of the system.
