← Projects · other

AI proposal generation for Gijima

Azure OpenAI (GPT-4) proposal-generation API for Gijima, with structured prompting, deterministic pricing, SharePoint as the system of record, and RAG contract analysis.

Built for Gijima, this service drafts complete client proposals from structured inputs (scope, services, pricing) and a corpus of prior proposals held in SharePoint Online. Authors edit a strong generated draft instead of starting from a blank page.

Design choices that keep it trustworthy:

  • The LLM never prices anything. Costs are computed deterministically from structured pricing data and injected into the prompt and the document template.
  • Structured prompting. A single role-and-process prompt (client analysis → problem definition → solution design) returns JSON-constrained output consumed programmatically; a parallel service rewrites individual sections on request.
  • Retrieval where it belongs. A LangChain + FAISS module analyses contract documents, while SharePoint (via Microsoft Graph) remains the system of record for industries, services, pricing, customers, and generated proposals.
  • Observable by default. Every generation logs prompt, duration, tokens, and estimated cost.