Jobsphere: AI-Powered Career Copilot Achieves 89% Cost Reduction with Multilingual Support for Government Employment Platforms

Government employment websites often present challenges for users, particularly regarding complex navigation, limited language support, and a lack of personalised guidance. To address these issues, Srihari R, Adarsha B V, and Mohammed Usman Hussain, along with colleagues from Presidency University, introduce JobSphere, an AI-powered career assistant designed to improve the user experience on the Punjab government’s PGRKAM employment platform. JobSphere utilises advanced Retrieval-Augmented Generation techniques and supports English, Hindi, and Punjabi, significantly broadening accessibility for a diverse user base. This innovative system achieves high factual accuracy and rapid response times while operating on readily available hardware, making it a remarkably cost-effective solution, and evaluation demonstrates a substantial improvement in usability compared to the existing platform, ultimately connecting more people with trusted government job opportunities.

JobSphere is a platform developed for use in Punjab, known as PGRKAM. Key innovations include voice-enabled functionality.

JobSphere, RAG and 4-bit Quantization for PGRKAM

The development of JobSphere marks a significant methodological advancement in delivering accessible employment resources for users of the Punjab government’s PGRKAM platform. Scientists engineered a system leveraging the Llama 3. 2 3B language model, deployed on readily available consumer-grade GPUs through a 4-bit quantization technique. This process reduces memory requirements, achieving an 89% cost reduction compared to cloud-based implementations while maintaining high performance. At its core, JobSphere utilizes a Retrieval-Augmented Generation (RAG) architecture, grounding responses in verified documents to minimize inaccuracies and enhance trustworthiness.

To facilitate comprehensive mock test preparation, the team automated test generation from previous papers, significantly reducing test creation time. This process incorporates Named Entity Recognition (NER), Part of Speech (POS) tagging, and syntactic parsing, achieving 85% accuracy in extracting relevant information. Real-time web scraping, crucial for maintaining up-to-date job listings, was implemented with techniques to bypass anti-bot technologies, including Selenium and Beautiful Soup. Resume parsing converts unstructured documents into structured profiles with 92% accuracy, utilizing contextualized word embeddings and machine learning classifiers.

The job recommendation system employs a two-stage architecture, translating embeddings and utilizing a gradient boosted tree ranking mechanism, and incorporates cosine similarity for semantic matching. Furthermore, the system leverages FastAPI for asynchronous APIs, Pydantic validation, JWT for authentication, React 18 for rendering, Vite for builds, and TailwindCSS for responsive design, ensuring a scalable and efficient operation. The back-end algorithms, based on established data structures like B-trees and hash tables, further contribute to the system’s performance and scalability.

JobSphere Delivers Accurate, Accessible Career Guidance

The development of JobSphere represents a significant advancement in accessibility for job seekers in Punjab, delivering an AI-powered career assistant designed to overcome limitations in existing government employment websites. A key achievement lies in the system’s efficient deployment, made possible through 4-bit quantization, which reduces implementation costs by 89% compared to cloud-based solutions while maintaining high performance. Experiments reveal JobSphere achieves 94% factual accuracy in its responses, demonstrating a robust ability to provide reliable information to users. The median response time measures at 1.

AI Assistant Improves Punjab Employment Access

JobSphere represents a significant advancement in government employment platforms, specifically designed to address accessibility and usability challenges faced by users in Punjab. Researchers developed an AI-powered career assistant that integrates data from the PGRKAM employment portal with large language models, creating a system capable of providing verified and relevant job information. A key innovation lies in its multilingual capability, supporting English, Hindi, and Punjabi, including voice-enabled interaction, which broadens access for diverse users. The team successfully deployed this system on consumer-grade hardware, utilizing a 4-bit quantization technique that dramatically reduces computational costs, making it 89% cheaper than cloud-based alternatives.

Evaluation demonstrates high factual accuracy, with 94% of responses verified, and a rapid median response time of 1. 8 seconds. User testing, measured by the System Usability Scale, reveals a 50% improvement in usability compared to the existing PGRKAM platform. While the current implementation demonstrates substantial improvements, the researchers acknowledge the need for continued refinement of the resume parsing capabilities and expansion of the mock test database. Future work will focus on enhancing these features and exploring the potential for personalized career guidance based on user profiles and skills.