Core Functionality:

The application leverages Retrieval-Augmented Generation (RAG) to provide accurate, cited answers from trusted internal sources. Its primary functions include:

• Data Ingestion: Supports manual uploads of PDF, HTML, and Text files, as well as an automated, scheduled web crawler to pull content from internal websites.
• Document Processing: Chunks documents into meaningful paragraphs, prepends contextual location data, and converts them into vector embeddings for storage.4
• Intelligent Retrieval: Performs vector searches to provide direct answers to user queries, complete with links to the exact document sections used as sources.
• Multimodal Interaction: Includes support for voice-to-text queries and audible voice output for hands-free use.
• User Management: Features a multi-tiered dashboard (Admin, Editor, Reader) with role-based access control.
• Asynchronous Scaling: A scalable system (v3) uses asynchronous job queues to handle large files and high-volume processing simultaneously.

Technical Stack

The project utilizes a robust modern technology stack:

• Backend: Python Django and Django REST Framework for the core API and application logic.
• Frontend: Next.js and React for the customer-facing dashboard and user interfaces.
• AI & LLM:
  • Models: GPT-4, GPT-4o, and GPT-4o-mini.
  • Embeddings: OpenAI text-embedding-3 (large and small).
  • OCR: PaddleOCR for multilingual text extraction.
• Data Storage:
  • Relational: PostgreSQL.
  • NoSQL: MongoDB.
  • Vector Database: pgvector within Postgres.
• Infrastructure (AWS):
  • Compute: Lambda and EC2
  • Storage: S3 for document and artifact storage.
  • Messaging: SQS, Redis, and Celery for task queuing and event handling.
  • Monitoring: CloudWatch for logging.
• Security: Keycloak (OIDC) for authentication and Role-Based Access Control (RBAC), PropelAuth for session management, and encryption at rest

The post AI-based project development first appeared on Codaemon Softwares Pvt Ltd.