InfoGraph AI

InforGraph AI boasts excellent response performance through a complex inference engine and the application of generative AI using LLM models. By utilizing data augmentation techniques and machine reading technology, it minimizes the need for extensive chatbot training data, allowing for quick and easy chatbot deployment. With this, implementing a Knowledge Graph-based ChatBot provides a high-performance customer complaint response manager available 24/7, helping to move away from handling repetitive customer complaints.

In traditional work systems, it has been very difficult for users to search for the information they need using only DB searches, tags, and Full-Text Retrieval (FTR). The reasons for this include dependency on keywords, lack of contextual understanding of questions, and limitations in natural language comprehension for complex and comprehensive questions.

So we introduce the emerging technological trends in information retrieval using LLMs and the solutions offered by BSG to these problems facing companies.

LLMs are great, but:

Knowledge cutoff date

Hallucinations

Lack of information source

Lack of domain-specific or private information

Supervised fine-tuning an LLM:

Generating training data is hard and expensive

Doesn't solve knowledge cutoff issue, only pushed it to a later date

No data sources

Hard to debug, is the information coming from foundation training set or from fine-tuning set or is it hallucinated?

Tunning is required when you want the model to learn something niche or specific that deviates from general language patterns. for example, you can use model tuning to teach the model the following:

Specific structures or formats for generating output.

Specific behaviors such as when to provide a terse or verbose output.

Specific customized outputs form specific types of inputs

Retrieval-augmented generation (RAG) is a technique for enhancing the accuracy and reliability of generative AI models with facts fetched from external sources.

In other words, it fills a gap in how LLMs work. Under the hood, LLMs are neural networks, typically measured by how many parameters they contain. An LLM’s parameters essentially represent the general patterns of how humans use words to form sentences. That deep understanding, sometimes called parameterized knowledge, makes LLMs useful in responding to general prompts at light speed. However, it does not serve users who want a deeper dive into a current or more specific topic.

Retrieval-augmented generatoin(in-context passing):

Providing relevant context in prompt

Avoid relying on facts provided by LLMs

Increased transparency due to provided sources

The answer is as good as the provided context

In a typical manufacturing-based company, the proportion of structured and unstructured data may vary, but a rough estimate can be made. In manufacturing, various types of data are generated, and these data are used in various areas such as production processes, quality control, supply chain management, maintenance, and customer service.

Structured Data

Structured data mainly exists in the following forms:

ERP system data: Data generated from an enterprise resource planning (ERP) system, including data from various areas such as finance, human resources, production, inventory, and order management.

MES system data: Data from a manufacturing execution system (MES), including real-time data generated from the production process, production schedules, and work instructions.

SCM system data: Data from a supply chain management (SCM) system, including data generated to manage each stage of the supply chain.

Quality control data: Data including quality inspection, test results, and defect reports.

Logistics and inventory data: Data including product movement, inventory levels, and warehouse management data.

Unstructured Data

Unstructured data mainly exists in the following forms:

Equipment and machine logs: Includes log data generated by machines and equipment, sensor data, maintenance records, etc.

Documents and reports: Includes technical documents, design drawings, project reports, manuals, etc.

Communication data: Includes emails, meeting minutes, chat logs, etc.

Images and videos: Includes quality inspection images, production process monitoring videos, etc.

Customer feedback: Includes customer service records, social media feedback, survey responses, etc.

When discussing the order in which a company should introduce a generative AI-based knowledge system to resolve employees’ asymmetric information access, increase the utilization of internal knowledge systems, and rapidly improve work efficiency and productivity, the characteristics of structured and unstructured data should be considered. In general, it is recommended to approach in the following order:

1. Utilization of Structured Data

Structured data is already organized systematically and is easy to analyze and process, so utilizing it in the early stages can quickly produce meaningful results.

1. Data purification and integration:
• Introduction of ETL process: Establish an ETL (Extract, Transform, Load) process for data purification and integration to integrate data from various sources.
• Data quality management: Establish quality management procedures to maintain the accuracy, consistency, and up-to-dateness of data.
2. Establishment and utilization of BI system:
• Data warehouse: Establish a data warehouse to store and manage structured data.
• Dashboard and report: Use BI tools to create real-time dashboards and reports so that employees can easily access them.
3. AI model training and deployment:
   • Machine learning model: Build predictive models, recommendation systems, etc. using structured data.
   • Automated report generation: Introduce an AI-based automatic report generation function to support rapid decision-making.

2. Leverage unstructured data

Unstructured data takes longer to analyze and process, but it contains a vast amount of useful information and can provide greater value in the long run.

1. Data collection and preprocessing:
   • Data ingestion: Collect data such as log files, emails, documents, and social media.
   • Preprocessing: Cleanse and structure the data using natural language processing (NLP), text mining, and image and video processing technologies.
2. Build a search and indexing system:
• Introduce a search engine: Use a search engine such as Elasticsearch to index unstructured data and enhance search capabilities.
• Knowledge graph: Build a knowledge graph to visualize relationships between data and provide better search results.
3. Introduce a generative AI model:
   • Natural language generation (NLG): Introduce an AI model (such as GPT) that generates natural language responses based on unstructured data.
   • Question-answering system: Build a system where employees input questions in natural language and AI provides appropriate responses.

3. Integration and Optimization

Build a comprehensive knowledge system by fusing structured and unstructured data and continuously improving it.

1. Build an integrated data platform:
• Hybrid data store: Build a hybrid data store that can store and manage both structured and unstructured data.
• Unified dashboard: Integrate various data to visualize and analyze it in a single dashboard.
2. Continuous learning and improvement:
• Collect user feedback: Continuously collect feedback from system users to improve the system.
• Model update: Periodically update the AI ​​model based on the latest data to maintain accuracy.
3. Education and culture creation:
• Employee education: Provide training programs so that employees can effectively use the new system.
• Data-driven culture: Build a culture that makes decisions based on data to increase the usability of the system.

In the early stages, it is efficient to use structured data to achieve results quickly and then move on to using unstructured data based on this. This step-by-step approach reduces the complexity that may arise during the system construction process and helps employees quickly adapt to the new system.

InfoGraph AI can solve problems with certainty and speed. In the manufacturing industry, it is a lump of various parts and processes. Unlike existing chatbots that simply provide defined answers, InfoGraph AI's knowledge graph-based chatbot utilizes the connection information linked to the data and provides answers. For example, if you enter the error code of a specific machine, it can immediately provide the cause of the error, the affected part, and the processing method.

InfoGraph AI also has a vitality data integration function. In winter, you can collect and manage data from many systems. InfoGraph AI integrates various data sources to display the whole picture. From the status of the production line to inventory management and maintenance records, you can process them all in one place. This enables more efficient decision-making.

Together, InfoGraph AI will also be of great help in maintenance. The chatbot can analyze the status of the machine and past data to predict and warn of problems in advance. This can especially prevent expensive machines and reduce maintenance costs.

And InfoGraph AI can provide information to users. It improves work capabilities by providing tailored information according to each user's role and needs. For example, field properties can easily obtain necessary technical support information, managers can easily obtain performance-related data, and supply managers can easily obtain inventory status and related information through chatbots.

Finally, InfoGraph AI leads to improved customer service. InfoGraph AI's chatbot solution operates 24/7 and can respond to customer inquiries. It provides customers with the information they want quickly, so you can exclude customers and save time only for repetitive inquiries.

BSG provides solutions for all information-related needs, from strategy development to system implementation and operation. With the increasing demand for transitioning from on-premises solutions to cloud environments, significant time is spent on staffing and onboarding, leading to business opportunity costs. To address this, BSG offers an environment where cloud-based solutions can be quickly validated and operated, ensuring swift onboarding in line with the AWS Marketplace launch.

BSG operates in an AWS-optimized cloud-native manner, ensuring high availability through VPC networks and EC2 auto-scaling groups, rapidly deploying Docker containers using boto3-based Python scripts to meet various customer requirements. Additionally, BSG plans to leverage AWS Marketplace to support metering and billing functions.