GENESIS is a multi-agent system designed to streamline the process of identifying and implementing AI/GenAI use cases tailored to specific industries and companies. The system operates through a series of specialized agents that perform individual tasks, gathering insights and generating solutions based on user inputs. Feedback is incorporated throughout the workflow to ensure relevant and practical outcomes.
- Objective: Gather the necessary details from the user to customize the GENESIS workflow.
- Inputs:
- Mandatory: Industry name and company name.
- Optional: Specific focus areas such as operations, supply chain, etc.
- Implementation:
- A simple web interface (e.g., Flask or Streamlit) or CLI can be used to collect this information.
- Inputs are validated through predefined dropdowns or regex to ensure correctness.
- Task: This agent performs retrieval tasks to gather company-specific and industry-specific insights.
- Subtasks:
- Search the web for company profiles, articles, or annual reports.
- Extract key insights such as company goals, policies, and focus areas.
- Retrieve policy documents or reports from websites or PDFs (with OCR if necessary).
- Tools:
- Web scraping: BeautifulSoup, Selenium, Scrapy.
- Document processing: PyPDF2, Tesseract OCR.
- Summarization: OpenAI API, Hugging Face transformers (e.g., BART, Pegasus).
- Knowledge retrieval: LangChain with vector databases like Pinecone or FAISS.
- Output:
- Key company insights and industry trends.
- Extracted policies or strategic documents.
- Task: This agent generates AI/GenAI use cases that align with the company’s strategic goals.
- Subtasks:
- Analyze the company’s goals, challenges, and industry trends.
- Generate a list of AI use cases that could help address these challenges (e.g., improving operations, enhancing customer experience).
- Tools:
- Language models: OpenAI API (GPT-3.5/4), Cohere, Anthropic Claude.
- Template-based generation: Use predefined templates that can be customized with AI-driven suggestions.
- Output:
- A set of AI/GenAI use cases with descriptions, linked to the company's needs.
- Task: This agent identifies and gathers resources required to implement the generated use cases.
- Subtasks:
- Search for relevant datasets from platforms like Kaggle or Hugging Face.
- Identify the required tools, libraries, and APIs needed for implementation.
- Generate clickable links to access these resources.
- Tools:
- APIs: Kaggle API, Hugging Face Datasets API, GitHub search API.
- Libraries: Requests for web queries, LangChain for semantic search.
- Automation: Python scripts to gather and present resource links.
- Output:
- A structured list of resources, such as datasets, libraries, and tools, necessary to implement the use cases.
- Task: This agent compiles all the collected information into a comprehensive, user-friendly report.
- Subtasks:
- Format the insights, use cases, and resources into a well-structured report.
- Add interactive elements such as hyperlinks to external resources for easy access.
- Tools:
- Report generation: Markdown or HTML templates.
- PDF export: ReportLab, WeasyPrint for generating downloadable reports.
- Output:
- A final report in PDF or HTML format, ready for presentation or further analysis.
- Web Scraping: BeautifulSoup, Scrapy, Selenium.
- Data Retrieval: LangChain, Pinecone, FAISS.
- Language Models: OpenAI GPT-3.5/4, Hugging Face models.
- APIs: Kaggle, Hugging Face Datasets, GitHub Search.
- Automation: Python libraries like Requests, PyPDF2.
- Frontend: Streamlit/Flask for user interaction.