Autonomous Amazon Order Details Fetcher -- 2
Budget: $30 – $250 USD
Autonomous agents capable of planning, reasoning, and executing actions on the web offer a promising avenue for automating computer tasks. Please build an autonomous agent that is capable of fetching user-level order details from the Amazon website.
Detailed Requirements
The agent should navigate to the "Order Details" page of each historical order of a particular user.
The agent should save the corresponding page as raw HTML files in a structured format (e.g., using a naming convention that includes the order number and date).
The agent should extract key order details from the HTML files, such as order number, product names, quantities, prices, and delivery status, and store them in a structured format (e.g., JSON or CSV).
The agent should make a decision to stop the process once all of the orders are fetched successfully.
Tips and Guidelines
The solution should leverage Large Language Models (LLMs) or Multimodal Language Models for natural language understanding, decision-making, and navigation. While other approaches can be combined with LLMs, the primary focus should be on utilizing LLMs effectively.
The agent should handle user authentication securely, ensuring that user credentials are protected and the authentication mechanism is reliable.
The agent should understand and interact with visual elements of a web page, such as buttons, forms, and navigation menus, to make informed decisions about the best actions to take.
The agent should be robust enough to handle unexpected events, such as pop-up windows or website changes, and adapt its behavior to ensure the task is completed successfully.
The solution should be designed with scalability and adaptability in mind, allowing it to be extended to different types of websites or web apps with minimal modifications.
If you encounter challenges in implementing the actual solution, you can discuss your thought process, including the proposed architecture, potential approaches, and any limitations or considerations.
Evaluation Criteria
Completeness and functionality of the solution, including the ability to handle authentication, navigate through order pages, extract relevant order details, and save them in a structured format. Bonus points for including production-ready architecture and design considerations.
Code quality, adhering to best practices, readability, and maintainability.
Effectiveness of data preprocessing techniques applied to the extracted order details.
Appropriateness and relevance of feature engineering techniques used, if applicable.
Model inference time and overall performance of the agent, considering the trade-off between accuracy and efficiency.
Application of optimization techniques, such as quantization, to improve the agent's performance. Evaluation metrics such as MMLU, TruthfulQA, and HumanEval accuracy for 4- bit and 8-bit implementations can be considered, if relevant to the chosen approach.
Utilization of open-source base models and effectiveness of fine-tuning strategies employed.
Quality and clarity of the submitted solution, including the results, model accuracy, and confusion matrix, if
Detailed Requirements
The agent should navigate to the "Order Details" page of each historical order of a particular user.
The agent should save the corresponding page as raw HTML files in a structured format (e.g., using a naming convention that includes the order number and date).
The agent should extract key order details from the HTML files, such as order number, product names, quantities, prices, and delivery status, and store them in a structured format (e.g., JSON or CSV).
The agent should make a decision to stop the process once all of the orders are fetched successfully.
Tips and Guidelines
The solution should leverage Large Language Models (LLMs) or Multimodal Language Models for natural language understanding, decision-making, and navigation. While other approaches can be combined with LLMs, the primary focus should be on utilizing LLMs effectively.
The agent should handle user authentication securely, ensuring that user credentials are protected and the authentication mechanism is reliable.
The agent should understand and interact with visual elements of a web page, such as buttons, forms, and navigation menus, to make informed decisions about the best actions to take.
The agent should be robust enough to handle unexpected events, such as pop-up windows or website changes, and adapt its behavior to ensure the task is completed successfully.
The solution should be designed with scalability and adaptability in mind, allowing it to be extended to different types of websites or web apps with minimal modifications.
If you encounter challenges in implementing the actual solution, you can discuss your thought process, including the proposed architecture, potential approaches, and any limitations or considerations.
Evaluation Criteria
Completeness and functionality of the solution, including the ability to handle authentication, navigate through order pages, extract relevant order details, and save them in a structured format. Bonus points for including production-ready architecture and design considerations.
Code quality, adhering to best practices, readability, and maintainability.
Effectiveness of data preprocessing techniques applied to the extracted order details.
Appropriateness and relevance of feature engineering techniques used, if applicable.
Model inference time and overall performance of the agent, considering the trade-off between accuracy and efficiency.
Application of optimization techniques, such as quantization, to improve the agent's performance. Evaluation metrics such as MMLU, TruthfulQA, and HumanEval accuracy for 4- bit and 8-bit implementations can be considered, if relevant to the chosen approach.
Utilization of open-source base models and effectiveness of fine-tuning strategies employed.
Quality and clarity of the submitted solution, including the results, model accuracy, and confusion matrix, if