Velotio Revolutionized a Manufacturing Data Platform with ChatGPT Integration Catalyzing 10X Productivity Surge

The project's foundation laid in utilizing ChatGPT's advanced analytical capabilities to tackle complex machine data challenges. As we progressed through the project, our focus was to ensure the seamless portability of ChatGPT into our client's existing product.

We achieved this step by step, in phases.

1: Data Generation and Tokenization

• Utilize SDKs to extract data from the platform.
• Apply tokenization techniques to ensure uniform data formatting.
• Generate initial "prompts" (questions) and their corresponding responses.
• The data load script runs periodically and captures any new data since the last run called as dataset.
• The important parameters like entity details, performance parameters are extracted from this dataset.
• The performance parameters are correlated with the factors affecting them (tags) to create descriptive sentences and summary sentences.
  Example of summary sentences:
  1st January 2023 the Mixes had Availability XXX and Running for 30 minutes.
  Performance, measured by TAG2015240, was all null.
  Quality, measured by TAG5653, averaged 30 units.
  Example of descriptive sentences:
  The Boiler machine is part of the packaging line. The following Boiler fields are highly correlated: TAG2541, TAG258242, TAG53842
• Such different types of sentences are passed to OpenAI embeddings to create a vector representation of the text called as Embedding.
  Example-
  "embedding": [
     -0.006929283495992422,
     -0.005336422007530928,
     ...
     -4.547132266452536e-05,
     -0.024047505110502243
    ],

2: Data Ingestion

• Ingest the generated data into vector or relational databases.
• Prepare the data for evaluating the effectiveness of text-to-query suggestions from OpenAI.

3: Choosing the Right Regression Algorithm

• Evaluate various regression algorithms such as OLS, RandomForest, etc.
• Finalize the most accurate regression algorithm (e.g., X) based on performance assessments.

4: Model Querying

• Parse and convert user input text into SQL queries.
• Execute these queries on the database to extract relevant data.

5: Accuracy Validation

• Iteratively evaluate OpenAI's text-to-SQL query performance.
• Validate the accuracy of the generated queries and make necessary refinements.

6: Portability

• Integrate the ChatGPT application back into the platform using REST APIs.
• Ensure seamless communication between ChatGPT and the existing product.
• Through these well-defined phases, the project aims to enhance the capabilities of the platform by incorporating ChatGPT's text-to-SQL query functionalities. The iterative evaluation and algorithm selection process ensure the delivery of accurate and effective results to users, fostering a user-friendly and productive experience.
• Finally, the integration of ChatGPT via REST APIs ensures the successful deployment of the application, empowering the platform to provide efficient and intelligent data querying solutions.

Here’s how the data flowed within the platform(post ChatGPT  integration):

Step 1- Data Load: Data is loaded as explained in the step-by-step process above.

Step 2 - User Response Flow:

2.1. The user enters their query in the chat box.
2.2. GPT converts the query to SQL. In this step, details such as entities, performance parameters, and associated fields are extracted to construct an SQL query.

Example- Question: "What factors affected the Availability for the Boiler machine today?"
In this case, the extracted information includes: machine = Boiler, performance parameter = Availability, date = 1st Jan 2023.

2.3. Vector Distance Calculation:
The extracted fields are used to query related data in the database.
The user's input query is converted into a vector embedding, and the distance between this vector embedding and the available vector embeddings in the database is calculated. The results are referred to as facts.
These facts are augmented with additional information, if necessary, to compose a coherent sentence-like response.

Overall, the provided content effectively outlines the steps involved in the user response flow, illustrating how ChatGPT interacts with the manufacturing data platform.