Skip to content

PRD.md

Latest commit

 

History

History
91 lines (46 loc) · 4.29 KB

PRD.md

File metadata and controls

91 lines (46 loc) · 4.29 KB

Product Requirement Document: Autonomous Operations Suite (Hackathon Prototype)

  1. Product Name: Autonomous Operations Suite

  2. Problem Statement: Retailers often struggle with sub-optimal inventory levels, reactive pricing strategies, and uncoordinated promotions, leading to waste, missed sales opportunities, and reduced profitability. Traditional, rule-based systems are too slow and inflexible to adapt to real-time market changes. This prototype aims to address these challenges by providing a proactive, intelligent system for retail optimisation.

  3. Product Goal (Hackathon Scope): To build a multi-agent Generative AI (GenAI) prototype that proactively optimises inventory, pricing, and promotions in real time to reduce waste and maximise profit within a simulated retail environment. The prototype should effectively demonstrate the collaboration between specialised AI agents.

  4. Target Users:

    Retail operations managers

    Store managers

    Merchandising teams

  5. Key Features & Functionality (Agent Roles & Responsibilities): The suite will consist of three collaborating specialised AI agents:

    5.1 Pricing Agent

     Core Responsibility: Dynamically adjust product prices to maximise profit and reduce waste.

 Key Inputs: Real-time demand elasticity, competitor prices, inventory levels.

 Key Outputs: Profit-weighted price moves, micro-discounts.

 Collaboration Point: Negotiates with the Inventory Agent to apply markdowns on at-risk stock.

    5.2 Inventory Agent

     Core Responsibility: Maintain optimal stock levels and trigger restocking.

 Key Inputs: Probabilistic demand forecasts, IoT shelf data (simulated), lead-time models.

 Key Outputs: Self-calibrating safety buffers, restocking alerts.

 Collaboration Point: Informs the Pricing Agent of slow-moving items to trigger price adjustments.

    5.3 Promotion Agent

     Core Responsibility: Orchestrate flash sales and create promotional bundles.

 Key Inputs: Social media sentiment (simulated), event schedules, SKU performance data.

 Key Outputs: On-the-fly bundle creation, pre-allocation of stock to fulfilment centers (simulated).

 Collaboration Point: Collaborates with the Pricing Agent to create micro-discounts and with the Inventory Agent to ensure stock availability.

  6. Architecture & Technical Considerations (Hackathon Focus):

    Architecture Type: Multi-agent system (Centralised/Orchestrator model is recommended for hackathon simplicity).

    Foundational Components:

     Large Language Model (LLM): The central reasoning engine for agent decision-making.

 Data and Memory Layer: A vector database to serve as agents' "persistent memory" for recalling past interactions, user preferences, and product information. This layer will leverage a small, high-quality, first-party dataset (simulated or representative).

 Execution Layer: APIs and tools that agents can access to perform tasks (e.g., updating simulated inventory, applying simulated price changes).

    Recommended Tools (Hackathon Stack):

     Generative AI Model: AWS Bedrock models

 Vector Database: ChromaDB

 Agent Framework: AWS Strands Agents

 Execution Layer: FastAPI or pre-built connectors.

 User Interface: Python/Streamlit, HTML/CSS/JavaScript for a simple web interface.

  7. Success Metrics (Hackathon Evaluation): The prototype will be considered successful if it can:

    Clearly demonstrate the collaborative interaction between the three specialised agents.

    Showcase how complex, real-world retail problems can be decomposed and solved by a team of collaborating bots.

    Visually represent how the system proactively optimises key retail metrics (e.g., reduces simulated stockouts, adjusts prices based on simulated demand).

    Provide a clear, value-led demonstration of GenAI's capability to drive efficiency and reduce waste in retail operations.

  8. Out of Scope (for Hackathon Prototype):

    Full-scale enterprise integration with live inventory, CRM, or POS systems.

    Extensive, real-world data foundation and data cleaning processes (focus on a representative dataset).

    Production-ready security, scalability, and robust error handling.

    Decentralised multi-agent architecture (will focus on a centralised orchestrator model).