Maya Knows

• Description: Maya Knows is an enterprise AI assistant that unifies internal knowledge from documents, tools, and systems. It enables natural language search and delivers intelligent summaries through a chat interface, designed to improve knowledge access across teams. The platform leverages advanced AI technologies to help organizations find information quickly, make better decisions, and enhance productivity by connecting disparate data sources and providing contextual insights. Built with modern web technologies and AI capabilities to create a seamless knowledge management experience for enterprise users.

• Project Overview:

  • Develop an enterprise-grade AI assistant for knowledge management and retrieval.
  • Implement natural language processing for intelligent search across multiple data sources.
  • Create a conversational chat interface for intuitive knowledge access.
  • Integrate with various enterprise tools, documents, and systems.
  • Build intelligent summarization capabilities for complex information.
  • Design scalable architecture to handle enterprise-level data volumes.
  • Implement secure authentication and authorization for enterprise users.
  • Deploy on cloud infrastructure with high availability and performance.

• Technologies Used:

  • React
  • Node.js
  • Python
  • FastAPI
  • PostgreSQL
  • OpenAI
  • LangChain
  • Vector Databases
  • AWS

• Role: As a Full Stack Engineer, I was responsible for developing both frontend and backend components, implementing AI integration, designing the knowledge management system, and ensuring seamless user experience across the platform.

• Client Requirement:

  • Create an enterprise AI assistant that unifies internal knowledge from multiple sources.
  • Implement natural language search capabilities for intuitive information discovery.
  • Develop a chat interface for delivering intelligent summaries and responses.
  • Ensure scalability and security for enterprise-level deployment.
  • Integrate with existing enterprise tools and document management systems.

• Workflow Drag-and-Drop Builder: Architected and developed a sophisticated visual workflow designer using pure vanilla JavaScript with advanced DOM manipulation and event handling. The system features an intuitive drag-and-drop interface where users can create complex knowledge processing workflows by connecting nodes representing different AI operations, data sources, and decision points. Each workflow is represented as a directed graph with nodes (processing units) and edges (data connections), stored optimally in PostgreSQL using adjacency list representation with JSON fields for node configurations. The builder supports:

  • Visual Node System: Custom-built node components for different operations (data ingestion, AI processing, filtering, transformation, output) with real-time connection validation

  • Data Flow Management: Implemented intelligent data routing system that tracks data loops, prevents circular dependencies, and optimizes execution paths through topological sorting algorithms

  • JSON Configuration Engine: Each workflow node stores its configuration as structured JSON in the database, enabling dynamic parameter management, conditional logic, and complex data transformations

  • Graph Database Optimization: Designed efficient database schema using PostgreSQL's JSONB columns for node properties and normalized edge tables for relationships, with indexed queries for fast workflow traversal and execution planning

  • Real-time Execution Engine: Built workflow execution system that processes nodes in optimal order, handles parallel execution branches, manages state persistence, and provides real-time progress tracking

  • Loop Detection & Prevention: Implemented sophisticated algorithms to detect potential infinite loops in workflow design, with automatic validation and user warnings during the design phase

  • Dynamic Data Binding: Created flexible system for binding data between nodes, supporting complex data transformations, conditional routing, and multi-output scenarios with type safety validation

Live project demo and proof of work is available on request!