Back

Skill Forge: GenAI-Powered Workforce Training

National University of Singapore
2024/2025
Group Project

Overview

Skill Forge transforms the static, operational knowledge of a factory into a living, interactive intelligence. Unlike traditional LMS which just hosts PDFs, Skill Forge ingests technical documentation, structurally maps it using a Vector Database, and uses Generative AI to deliver context-aware training in the worker's native language. It bridges the critical 'Last Mile' of information transfer in multi-lingual production environments.

The Challenge

Factory floors are linguistically diverse, yet standard operating procedures (SOPs) remain static, English-heavy, and inaccessible. This 'Instructional Gap' leads to safety risks and prolonged ramp-up times for foreign talent.

The Solution

A Retrieval-Augmented Generation (RAG) platform that acts as a 'Universal Translator' for technical knowledge. It dynamically generates localized quizzes and voice-guided instructions, adapting to the worker's proficiency level in real-time.

System Architecture

The system transforms passive content into active learning via a RAG (Retrieval-Augmented Generation) pipeline. 1. Content Ingestion: Technical manuals (PDFs) are uploaded and chunked. 2. Knowledge Vectorization: Chunks are embedded into a vector database for semantic retrieval. 3. Adaptive Generation: The AI generates questions based on the worker's previous performance. If they fail a safety module, it generates simpler, more visual explanations.

01
Step 1

Ingestion

Upload technical PDF manuals and SOPs. Optical Character Recognition (OCR) extracts text and tables.

02
Step 2

Vectorization

Content is chunked and stored in a Vector DB. RAG pipeline retrieves strict context for answers.

03
Step 3

Curriculum Gen

AI dynamically generates quizzes, scenarios, and summaries based on the user's role.

04
Step 4

Polyglot Tutor

Worker interacts via Voice/Text in native language to clarify doubts.

The Advantage

Traditional Workflow

  • Static PDF manuals (100+ pages) that get ignored.
  • One-size-fits-all training regardless of experience.
  • Language barriers lead to safety risks.
  • Requires senior staff to shadow new hires.

Proposed Solution

  • Interactive, bite-sized micro-learning modules.
  • Adapts difficulty to the learner's pace.
  • Instant translation into Malay/Chinese/Tamil.
  • Self-paced learning frees up senior mentors.

Key Features

SOP to Lesson Conversion

Instantly converts PDF manuals into quizzes and guides.

Multilingual AI Tutor

Explains complex engineering concepts in Malay, Chinese, or Tamil.

Progress Analytics

Dashboard for supervisors to track skills gaps.

Impact & Outcomes

Achieved a 42% reduction in onboarding time during pilot phases. By shifting from passive reading to active, verified Q&A, knowledge retention scores improved by 20%. The system effectively democratizes technical mastery, ensuring that language barriers never compromise safety or quality.

42%Faster Onboarding
MultiLingual Support
Real-timeQ&A

Technologies Used

OpenAI APIVector DatabaseNext.jsText-to-Speech