FM

Find A Masters AI Search Visualiser

Agentic RAG, pairwise ranking, and citation survival for postgraduate search.

Live Execution Pipeline

The point is the workflow: plan, route, gather, critique, compare.
Idle
1Plannerdecomposes prompt
2Routerchooses sources
3Toolsfetches evidence
4Criticspots gaps
5Pairwisesurvival round
pipeline.log agentic-rag
// Choose a query and run the simulation.
Step 1

Orchestrator plan

  • Run the simulation to see the decomposed query roadmap.
Step 2

What the retrieval finds

  • Retrieval findings will populate here.
Step 3

Self-critique verdict

The critic checks whether the evidence is specific enough to answer the user.
Pairwise Ranking

The content knockout round

The model does not just score pages in isolation. It compares candidate passages head-to-head, keeps the useful evidence, and drops generic filler.

Ready to play

Referee view
Competitor A

No active match.

Entities: 0Fluff: 0%
Competitor B

No active match.

Entities: 0Fluff: 0%
Micro-verdict
The model's reasoning appears here after each comparison.
Multi-Intent AI Search

Parallel brackets feed the final answer

A single user question can trigger multiple sub-queries. To win the final AI Overview, the brand needs evidence across the academic, practical, and experiential parts of the journey.

Grand Consensus Output

Waiting for winners

Qualified evidence

  • No bracket winners yet.

Run the consensus simulation to see how winning passages become a synthetic AI Overview answer.

Classic RAG

Static retrieval

A single query is vectorised, a few nearby documents are fetched, and the model writes from that one pass.

  • Weak on multi-intent questions.
  • No source routing to forums, video, or specialised databases.
  • No reflection loop to notice missing funding, cost, or lived-experience evidence.
  • Passages are treated as isolated matches, so generic copy can slip through.
Agentic RAG

Iterative research system

The system plans, routes to the right evidence source, critiques the first draft, then selects passages through comparison.

  • Breaks the user question into micro-queries.
  • Routes academic facts, social proof, and video evidence to different tools.
  • Self-corrects when the evidence is thin or contradictory.
  • Rewards pages with specific entities, direct answers, and useful supporting assets.
Meeting takeaway

What this means for content

1Write answer-first passages.

Every section needs to survive a head-to-head comparison, not just fill a page.

2Build evidence across surfaces.

Guides, course pages, video, and community proof all enter different retrieval routes.

3Own factual entities.

Precise figures, course types, requirements, and citations increase the chance of being selected.