VendorIQ — Multi-Agent Vendor Risk & Due-Diligence Platform
Home → Projects → VendorIQ
AI AGENTS · VENDOR RISK · IBM WATSONX
I built VendorIQ, a multi-agent vendor risk and due-diligence platform that investigates a vendor across five risk dimensions and combines the findings into a structured recommendation: Approve, Review, or Reject. Built for the IBM watsonx challenge, the system combines source-grounded evidence, confidence scores, and red-flag detection to support more transparent procurement decisions.
VendorIQ is designed to automate the first pass of vendor due diligence using publicly available information, specialised AI agents, and explicit decision rules. Its recommendations are intended to support human review, not replace it.
System
Multi-agent AI
Stack
Python · IBM watsonx · LangChain · Llama 3.3 70B
Focus
Vendor risk and procurement decision support
The Problem
Vendor due diligence can require reviewing financial health, regulatory compliance, operational capability, reputation, and strategic fit. Gathering this information from multiple sources and translating it into a consistent recommendation can be time-consuming and difficult to audit. I focused on structuring this process into a single workflow that brings together evidence across these dimensions while keeping the reasoning behind the final recommendation visible.
What I Did
I built the workflow around five specialised domain agents, each responsible for a different aspect of vendor assessment. I designed a public-data collection layer that gathers available information, then the agents analyse that evidence before a coordinating meta-agent combines their findings into a final assessment. I implemented evidence-grounded scoring so each domain contributes structured findings, confidence, supporting evidence, and potential red flags while keeping the reasoning behind the recommendation visible.
Financial
Evaluates liquidity, solvency, profitability, revenue stability, and credit indicators.
Compliance
Checks relevant certifications, sanctions information, and regulatory actions.
Operational
Examines available information about scale, continuity, and delivery capability.
Reputation
Reviews news signals, track record, and relevant certifications.
Strategic Fit
Assesses alignment with client priorities, dealbreakers, and preferences.
I implemented a two-pass agent design that separates evidence-based reasoning from structured scoring. Each agent produces a score out of 100 alongside confidence, supporting evidence, and potential red flags. The meta-agent combines the results using configurable weights and generates a plain-English recommendation with suggested actions.
Impact
VendorIQ brings multiple vendor-risk dimensions into one structured assessment workflow. By combining specialised agents with source-linked evidence, confidence scores, configurable weighting, and explicit compliance guardrails, the platform is designed to make initial vendor screening easier to review and its recommendations more auditable.
What it Demonstrates
VendorIQ demonstrates how agentic AI can be applied to enterprise risk workflows. The project brings together parallel agent execution, live-data collection, structured LLM outputs, weighted scoring, and deterministic rules to support a consequential decision-making process.
Python · IBM watsonx · LangChain · Llama 3.3 70B · yfinance · BeautifulSoup · asyncio