QuoteGuard_AI

Inspiration

I was inspired by the challenges of converting unstructured Requests for Quotation (RFQs) into accurate, reliable quotes. Manual procurement workflows can be time-consuming, and overlooking a critical detail—such as a conflicting material grade, missing certification, or unverified specification—can lead to costly mistakes. I wanted to build an AI-powered assistant that makes this process faster without sacrificing accuracy or trust.

What it does

QuoteGuard AI transforms unstructured RFQs into structured requirements, identifies missing information and conflicting specifications, matches requests against a product catalog, and prepares quote drafts for human review.

Instead of guessing when requirements conflict, QuoteGuard flags the issue and withholds the quote draft. For example, when an RFQ specifies both 316L and 304 stainless steel, QuoteGuard detects the conflict and requests clarification rather than proceeding with an unsafe match.

I also integrated PRISM for trace observability and guardrail monitoring, providing visibility into the analysis workflow.

How we built it

I built QuoteGuard AI using React and TypeScript for the frontend and Node.js/Express for the backend. The analysis workflow combines deterministic validation and catalog matching with an AI integration through Nicrron. PRISM provides trace ingestion and observability.

I designed the workflow around human oversight: the system structures information and suggests catalog matches, while incomplete or conflicting requirements must be resolved before a quote can proceed.

Challenges we ran into

One of my biggest challenges was handling conflicting specifications safely. My initial logic recognized stainless steel grades but did not reliably detect when an RFQ contained both 316L and 304. I updated the deterministic analysis logic to identify this conflict, mark the grade as requiring review, and return no catalog matches or quote draft.

I also worked through backend configuration, API integration, trace verification, and GitHub deployment under a tight deadline. These challenges reinforced the importance of testing failure cases, not just demonstrating successful workflows.

Accomplishments that we're proud of

I'm proud to have built a working end-to-end prototype that transforms RFQs into structured information, identifies missing requirements, and supports catalog-based quote drafting.

One particularly meaningful accomplishment was fixing the material-grade conflict detection. QuoteGuard now recognizes the tested 316L-versus-304 conflict and requests clarification instead of generating an unsupported quote. I also successfully connected PRISM trace ingestion and verified that live application analyses appear in the tracing dashboard.

What we learned

I learned that building a useful AI procurement tool requires more than extracting fields or generating text. Reliable automation depends on explicit validation, transparent evidence, observability, and clear boundaries around what the system can safely conclude.

I also learned the value of testing adversarial inputs and edge cases early. A system should not only perform well when an RFQ is complete; it must also recognize when it lacks enough reliable information to proceed.

What's next for QuoteGuard_AI

My next steps are to expand RFQ test coverage, improve conflict detection, validate supplier data and pricing, and further verify guardrail enforcement throughout the live workflow.

I envision QuoteGuard AI becoming a trustworthy procurement assistant that helps teams move from unstructured requests to review-ready quotes faster and more consistently, while keeping human judgment at the center of consequential quoting decisions.

Built With

• blockconvey
• gpt
• nicrron