Abstract

This publication introduces Lexica, an AI-driven virtual assistant designed to enhance accessibility and retrieval of corporate policies and internal documentation. Lexica leverages state-of-the-art natural language processing (NLP) models from Hugging Face or openai api to provide employees with real-time, contextually relevant responses to queries. This work explores the methodology behind Lexica’s development, evaluates its effectiveness through experimental testing, and discusses its impact on corporate knowledge management.

Introduction

In modern enterprises, employees frequently encounter difficulties accessing and understanding internal policies, such as leave regulations, reimbursement processes, and compliance requirements. Conventional document management systems often require manual searches, leading to inefficiencies and lost productivity.

Lexica addresses this challenge by introducing an intelligent, conversational interface that allows employees to interact with internal documentation naturally. By integrating AI-powered retrieval mechanisms, Lexica improves response accuracy, enhances policy adherence, and optimizes knowledge-sharing within organizations.

Methodology

Lexica is built on internal corporate documents. The system architecture comprises the following components:

• Data Preprocessing: Documents are parsed, structured, and embedded using vector representations in a ChromaDB storage system.

• Conversational AI: A transformer model (GPT-4 Mini) processes user queries and matches them with relevant document excerpts.

• Recommendation System: Role-based access control and personalization algorithms ensure tailored responses like email writing.

• Deployment and Scalability: The backend is hosted on Google Cloud, utilizing Flask for API management and containerized deployment.

Experiments

To evaluate Lexica’s efficiency, we conducted an experiment involving a dataset of 100 corporate queries across various domains (HR, finance, legal). The model’s response accuracy and retrieval speed were compared against a traditional keyword-based search system. Key performance indicators included:

• Response Accuracy: Measured as the percentage of correctly retrieved answers verified by HR experts.

• Retrieval Speed: Average response time per query.

• User Satisfaction: Assessed via survey feedback from employees interacting with Lexica.

Results

The experimental results demonstrate a significant improvement over traditional search methods:

Response Accuracy: 92% (Lexica) vs. 75% (Keyword Search)

Retrieval Speed: 1.2 seconds (Lexica) vs. 56.8 seconds (Keyword Search)

User Satisfaction: 88% of employees reported improved accessibility to policies and higher confidence in responses received.

These findings highlight the effectiveness of AI-driven assistants in streamlining corporate knowledge retrieval.

Conclusion

Lexica represents a powerful advancement in enterprise AI applications, improving internal policy accessibility and user experience. By integrating cutting-edge NLP models with intelligent document retrieval, Lexica significantly enhances organizational efficiency. Future research will focus on expanding multilingual support and integrating real-time feedback loops for continuous learning.