Skip to content
Mark Warren
Menu

All work

RFP response platform

An RFP response platform on Azure AI Search

An AI-assisted platform that drafts responses to new requests for proposals from a library of previously approved answers, built on Azure for a contractor in a regulated industry.

How it fits together

  1. Inputs

    • Past proposals

      Historical proposals and approved answers held in SharePoint.

      • SharePoint

    • A new RFP

      The proposal team drops the new request into a document library.

      • SharePoint

  2. Engine

    Response platform

    A Next.js application on Azure services that turns a new RFP into a first draft for review.

    • Azure AI Search
    • Next.js
    1. Extract the questions

      Document Intelligence reads the request from Blob Storage and pulls out each question with complex tables kept intact.

      • Document Intelligence
      • Azure Blob Storage
    2. Find approved answers

      Azure AI Search finds previously approved answers by meaning, not only by keyword.

      • Azure AI Search
    3. Draft from what was approved

      An OpenAI model assembles the response from the retrieved answers, so drafts start from wording the company already stood behind.

      • OpenAI

  3. Outputs

    • Branded draft

      A formatted Word draft goes to the proposal team for review.

    • People finish it

      The team refines, prices, and approves every response before it is submitted.

Guardrails

  • Nothing is submitted without human review
  • Drafts draw on previously approved answers
  • Documents stay inside the company's own Microsoft cloud environment

The problem

Each new RFP asked questions the team had already answered somewhere in past proposals. Finding those answers meant searching document libraries by hand, proposal after proposal, and that search, not the writing, limited how many bids the company could pursue.

What I built

  • Question extraction with Azure Document Intelligence that keeps complex tables intact, the detail that breaks most document automation.
  • A search index in Azure AI Search over the company's approved answers, with source documents in Azure Blob Storage and SharePoint history as the starting library.
  • Retrieval-grounded drafting with an OpenAI model, so each draft starts from approved wording instead of a blank page.
  • The application itself in Next.js, with people making every final call on what is submitted.

How I built it

I built it as technical lead at a former employer, a contractor in a regulated industry, alongside a retrieval-based help assistant for one of its applications. The design kept the company's documents inside its own Microsoft cloud environment and used Azure services for each stage: Blob Storage for documents, Document Intelligence for extraction, AI Search for retrieval, and an OpenAI model for drafting.

Manual searching dropped out of the proposal workflow, and the team's time moved from hunting for old answers to refining responses and pricing. Bidding capacity grew without adding proposal staff.

What it shows

I build retrieval systems on Azure for regulated work, where the useful answer is the one the company has already approved and a person still signs off.