Saurav Tripathy

AI Projects

MarTech Vendor Evaluation Agent

AI research tool evaluating MarTech vendors against a pre-determined scorecard.

Agentic Design

Researcher-Judge Workflow

Tech Stack

Python · LangGraph · Anthropic API (researcher) · OpenAI API (judge) · Gradio

How it works

STEP 1

Gather evidence

  • You give the agent a vendor name, your use case, and any documents you already hold, such as analyst reports or an RFP response. NOTE: Only vendor name is mandatory
  • The agent researches the vendor across public sources including product pages, analyst coverage, customer reviews, and trust portals
  • Every fact is mapped to the scorecard question it answers and graded by source strength, from vendor-stated to independently verifiable

STEP 2

Generate scores

  • The agent scores the vendor 1–5 (1=low, 5=high) on 28 weighted criteria across 8 categories
  • Categories include functional fit, integration, AI capability, security, cost, vendor viability, lock-in risk, and implementation support
  • Claims backed only by vendor marketing are capped at 3. Any missing evidence stays blank and is not guessed

STEP 3

Cross-examine & report

  • A second AI from a different provider plays judge and reviews the evidence gathered by the first agent
  • It challenges scores that are weakly supported, flags evidence gaps, and recommends adjustments
  • The final output is a downloadable Excel report with scores, sources, and judge dissent