AZURE AI FOUNDRY · RAG · AGENTS

Tender & Product Bot

One AI assistant, two jobs: instant product knowledge for everyone, and human-in-the-loop tender creation for Sales. Tender time cut from a month to two weeks.

Role

Product Owner & AI Lead

Timeline

2025 · shipped to Sales, PM & PMM

The challenge

Product knowledge, specs, prices, features, lived scattered across documents most people didn't know existed. Finding an answer took hours; building a customer tender took a month of hunting, assembling and checking.

The decision

Chose RAG on Azure AI Foundry over Fine-tuned model

Product data changes weekly, ground the model on live sources and answers stay current; fine-tune and you're retraining forever.

How I approached it

  • 01Collected use cases from Sales, PM and PMM, and split the product in two: open product knowledge for all, tender creation gated to Sales via Azure AD.
  • 02Wrote the PRD with the human-in-the-loop boundary explicit, the AI drafts the tender, the salesperson reviews, decides and owns what the customer sees.
  • 03Built RAG grounded on the product portfolio, specs, pricing and documentation, so answers cite the source instead of guessing.
  • 04Shipped through epics and user stories with adoption tracked from day one across the three user groups.

Impact

  • End-to-end tender creation from 1 month to 2 weeks.
  • Product answers from hours of document-hunting to seconds.
  • Consistent answers across a 5,000+ SKU portfolio.
  • Sales, PM and PMM working from the same source of truth.

Stack

  • Azure AI Foundry
  • RAG
  • Azure AD
  • Python