Introducing AI Smart Validator: AI-powered tool for validating SAP warehouse records using open-source LLMs and vector-based domain-specific knowledge. The idea emerged after projects in the SAP WM space earlier this year.

We all know it, but it’s worth reiterating: AI is only as good (or bad) as the data it’s being fed. LLMs are brilliant but generalists, while every business has its own competitive edge and its own magic. What we are trying to achieve is embedding Continuous Learning (the ‘Feedback’ reflects it), and use it as we go. It’s not a new idea, but more often neglected than implemented (does ChatGPT learn from your valuable input? The only honest answer is: we don’t know).

AI Smart Validator – What It Does

AI Smart Validator allows SAP teams to validate logistics data using large language models (LLMs) such as Mistral 7B, enhanced with domain logic and RAG capabilities. Connected with SAP backend, it can exchange and update data for best logistics process outcomes in real-time.

Features

• Validation Modes
  • Basic: Pure LLM reasoning on structured input
  • RAG (simplified for demo purposes): LLM + context from SAP-specific rules and warehouse documentation (via vector DB)
• LLM Output
  • Status per record (OK / Error)
  • Score per record (1–10 indicating data quality)
  • Detailed reasoning and explanation
• Summary View
  • Displays error counts, average score, execution details, and highlights
• User Actions (demo mode only)
  For each record, user can:
  • Accept: Confirms the AI decision (Accepted).
  • Reject: Cancels the validation result (Rejected).
  • Retry: Re-runs validation for particular item (can be done after sensitivity change, or without).
  • Email: Opens a dialog box to send email with auto filled item content.
  • Feedback: Allows users to submit feedback used to improve future validations (fine tuning / ML).
  • Worklist: Adds the item to the user Worklist for follow-up or escalation (Added to Worklist).
  • Sync to SAP: Sends the validated result to SAP backend for update (not connected in demo mode).

📦 Architecture

• Frontend: React + Tailwind CSS; deployed to Vercel and Netlify
• Backend: FastAPI (Uvicorn) with Groq (LLaMA 3 / Mistral); hosted on Render
• Data Flow: JSON input → LLM Reasoning → Results rendered per record
• RAG Engine: Vector-based Retrieval-Augmented Generation using ChromaDB

Integration Options

• Can be embedded or extended for:
  • SAP ECC
  • SAP BTP (CAP-based APIs)
  • SAP AI Core (doesn’t require S/4HANA)
• System-agnostic — while optimized for SAP, can be adapted to other platforms.

Try Me Out

• Smart Validator via React
• Smart Validator via Streamlit (simplified)

Contribute

💬 Got feedback? Feel free to contribute on GitHub, and if you’ve got suggestions about how to improve the App, share them in the comments or send a message. Best ideas will be rewarded!