Skip to content
Zester
All products

Procurement Automation

RFQ Agent

A multi-agent procurement system that watches public tender sources, scores every opportunity against fit, and drafts the proposal.

Domain
Business Operations
Status
In active development
Stack
LangGraphLangChainOpenAIFastAPIPostgreSQLNext.js 14TanStack Query

0–100

LLM fit score per opportunity

The problem

Most tenders are lost before anyone reads them.

Public procurement opportunities are published across separate portals on their own schedules. A team either assigns someone to check all of them daily, or accepts that it will miss the ones it was best positioned to win. The qualifying work — reading the requirement, judging fit, starting a draft — is the same every time and is exactly the work that gets skipped when the queue is long.

How it works

  1. 01

    Monitor the real sources

    CanadaBuys, SAM.gov, and MERX are watched continuously by scheduled collection rather than checked manually.

  2. 02

    Score against fit

    Each opportunity is matched by an LLM and scored 0 to 100, so the queue arrives ordered by whether it is worth the team's time.

  3. 03

    Draft the response

    Coordinated agents build a first proposal draft from the requirement, which turns qualification into editing.

  4. 04

    Alert where the team already is

    Scored opportunities are pushed to Slack, so the signal arrives in the working channel rather than in another dashboard.

Architecture

  1. 01

    Sources

    CanadaBuys, SAM.gov, MERX tender feeds

  2. 02

    Collection

    APScheduler-driven monitoring into PostgreSQL

  3. 03

    Scoring

    LLM matching producing a 0–100 fit score per opportunity

  4. 04

    Agents

    LangGraph and LangChain orchestration over OpenAI models for drafting

  5. 05

    Surface

    Next.js 14 with TanStack Query, FastAPI service, Slack alerts

What changes

  • The team sees every relevant tender rather than the ones it had time to check.

    Three sources are monitored on a schedule; coverage is inspectable in the pipeline.

  • Qualification becomes a review step instead of a research step.

    Every opportunity carries a 0–100 fit score before a human opens it.

  • Proposal work starts from a draft rather than a blank document.

    PILOT MEASUREMENT REQUIRED: hours from tender publication to first submitted draft.

Capabilities

  • Continuous monitoring of CanadaBuys, SAM.gov, and MERX
  • LLM opportunity matching with 0–100 fit scoring
  • Multi-agent proposal drafting
  • Scheduled collection via APScheduler
  • Slack alerting into the working channel
  • PostgreSQL-backed opportunity pipeline

Who uses it

Bid and proposal teams

Deciding which of this week's tenders is worth a response.

Business development leads

Seeing the pipeline before a competitor has already answered.

Capture managers

Starting a draft from a scored requirement rather than a blank page.

Evidence pending

This page does not fabricate product imagery. The slots below are reserved at their final dimensions and name exactly what is needed.

Product screenshot required

Opportunity pipeline with fit scores visible.

Sample output required

One scored opportunity and its generated draft, redacted.

Have an operation this should be pointed at?

Solve a Real Problem