Petronas - PetroNet, an AI-Powered Cognitive Search Solution

An AI-powered search platform that harnesses the power of Azure and NLP to revolutionise and streamline the way you navigate and unlock key insights within PETRONAS reports

Comment

Inspiration

Tired of spending hours sifting through complex and lengthy PETRONAS reports to find the information you need? Whether you are a prospective investor trying to learn more about PETRONAS, or an internal PETRONAS user looking to access specific information quickly, PetroNet offers an innovative solution for efficient information retrieval and index searching.

Solution

PetroNet is a one-stop AI-powered search web app that harnesses the power of Microsoft Azure services and Natural Language Processing (NLP) to revolutionise the way you unlock key insights within PETRONAS reports.

Key Functionalities

  1. Intelligent Information Extraction: PetroNet utilizes cutting-edge AI technologies to automatically extract and categorize text and images from PETRONAS reports, including Integrated & Annual Reports, Financial Reports and Sustainability reports. By employing Azure Cognitive Search skills such as Optical Character Recognition (OCR) and advanced text analytics, PetroNet captures the essence of each report, ensuring that no valuable insights are missed.
  2. Key Topics Identification and Categorisation: PetroNet leverages large language modelling capabilities of the Generative Pre-trained Transformer (GPT), specifically, the GPT-3.5 turbo model, to conduct topic modelling. With multiple prompts, the GPT analyses the content of PETRONAS reports to identify key topics and categorise them for easy navigation. PetroNet's intelligent tagging system organizes the reports into relevant themes, enabling users to quickly locate the information they are interested in.
  3. Powerful Search Capability: PetroNet's user-friendly landing page features a robust search bar enhanced with Azure Cognitive Search’s Natural Language Understanding (NLU) capabilities. Simply enter your query, and PetroNet will search across multiple reports, highlighting specific keywords in the content. This saves you precious time and effort, ensuring you find the most relevant information instantly.
  4. Comprehensive Results: PetroNet not only surfaces *relevant documents *related to your search query but also provides a *concise summary *of the relevant results within each document, streamlining your research process.
  5. Interactive Knowledge Graph: To help users better understand the context of their topics of interest, PetroNet generates a dynamic visual representation of relevant entities in a knowledge graph. This graph showcases the relationships between different entities, giving you a holistic view of the interconnected information within PETRONAS reports.

Technologies

I used Python, Azure Blob Storage, Azure Functions, Azure Cognitive Service and Azure Cognitive Search for the backend and C# .NET for the frontend to build the web application.

Challenges and Workarounds

  • Overcoming the limitations of the Cognitive Search Basic tier:
    • Implement workarounds to handle large PDF sizes by utilizing PyPDF2 to chunk up the PDF files and reduce their size for processing within the Basic tier.
  • Managing the limitations of the OpenAI API free tier:
    • Address token limit constraints by dividing large sections of text into smaller chunks that fit within the token limit, ensuring complete processing of the content.
    • Mitigate rate limit challenges by incorporating buffer time between every three API requests, optimizing the usage of the free tier resources.
  • Resolving dependency issues in the Azure function:
    • Actively troubleshoot and attempt to systematically debug dependency conflicts or compatibility issues encountered within the Azure function

Potential impact

With PetroNet, investors can gain valuable insights into PETRONAS' financial performance, market trends, and growth opportunities, while internal PETRONAS users can access relevant data for decision-making, project planning, and performance analysis.

What's next - Scale Up PetroNet for Enhanced Functionality

Our next step for PetroNet is to scale up its capabilities by transitioning to a higher-level cognitive search service tier. Currently, I am utilizing the Basic tier (~$75.14/month) to optimize costs using my free Azure credits. However, upgrading to a tier like Standard S1 (~$240/month) will unlock a wider range of functionalities. With this upgrade, PetroNet will be able to index a larger number of report pages, including the original-sized reports, and handle a higher volume of documents.

One of the key benefits of the upgraded tier is the incorporation of semantic search, a powerful feature that enhances search results with semantic relevance and language understanding. By leveraging semantic search, PetroNet will deliver more accurate and contextually relevant search results, further improving the efficiency and effectiveness of information retrieval.

Go-to-Market

  1. Integration with PETRONAS Reports Page Collaborate with PETRONAS to integrate PetroNet into their official PETRONAS Reports webpage or create a dedicated section for PetroNet. This integration ensures maximum visibility and accessibility to the target users, allowing them to easily access PetroNet's search capabilities and insights while browsing PETRONAS reports.
  2. Adoption Campaign Conduct a comprehensive marketing and adoption campaign to raise awareness about PetroNet's capabilities among the target users. Collaborate with PETRONAS to promote PetroNet through various channels, including PETRONAS' official communications, social media, newsletters, and industry events, highlighting the benefits of PetroNet, such as time savings, enhanced insights, and improved decision-making, to attract and engage users.

Built With

Share this project:

Updates