Skip to content

Problem_Definition.md

Latest commit

 

History

History
24 lines (23 loc) · 2.5 KB

Problem_Definition.md

File metadata and controls

24 lines (23 loc) · 2.5 KB
Raw

Problem Definition & Design Brief

  1. Context & Problem Statement
  • Domain: Consumer Goods (FMCG) – Beverages Category.
  • Current State: Business users frequently require insights regarding promotional performance summaries, inventory movements, regional sales comparisons, and product-level campaign impacts.
  • Pain Points: Currently, these analytical requests are managed via manual dashboarding and ad-hoc analysis by the Data & AI team. This creates a bottleneck, leading to delayed decision-making, repetitive manual tasks, and an over-reliance on data analysts.
  1. Project Objective
  • Primary Goal: To design, build, and deploy an AI-powered conversational assistant capable of answering business users' analytical questions autonomously.
  • Outcome: Empower business users to obtain immediate, accurate insights conversationally, thereby eliminating the need to involve the data team for routine or ad-hoc data requests.
  1. Proposed Solution & Scope
  • Architecture Pattern: A Text-to-SQL AI Agent system.
  • Data Scope: The assistant will reason over a synthetic relational database structured specifically for FMCG use cases. The core data model will include: o Sales & Promotions: Weekly sales performance and promotional activity. o Inventory: Weekly store-product level stock movements, enabling tracking of stockouts and over-stocking. o Product Master: Reference data for beverages (brands, categories, pricing). o Store Master: Reference data for retail locations (regions, store formats).
  1. Key Design Principles & Constraints
  • Autonomy in Infrastructure: The environment setup, data generation, tooling selection, and infrastructure deployment are completely managed from scratch.
  • AI-Assisted Development: AI tools (such as ChatGPT, Claude, and Supabase) will be actively utilized throughout the ideation, coding, and data generation phases to maximize efficiency.
  • Accuracy & Hallucination Mitigation: Because the system handles business-critical financial and inventory data, the design must prioritize strict data reasoning. The LLM will be restricted to generating read-only SQL queries rather than attempting to calculate math internally, mitigating the risk of hallucinations.
  1. Success Criteria (Evaluation Alignment)
  • Problem Framing: Clear decomposition of the user intent and effective use of prompt engineering.
  • Solution Design: Appropriate use of agentic patterns to connect natural language to the database.
  • Implementation: Delivering a working, end-to-end prototype with clean logic.