Methodology
How DisputeIQ turns a plain-English question into cited research
A retrieval-plus-generation pipeline designed so the AI is never the source of truth for official facts — it only explains what the retrieved decisions actually say.
Interpret the query
Your natural-language description is turned into semantic concepts and any explicit filters — without inventing a desired outcome.
Retrieve broadly
A hybrid search over concise decision summaries pulls a wide candidate set, with your filters applied before ranking.
Rerank against your query
Candidates are reranked on factual, product, conduct and reasoning similarity — not just keyword overlap.
Deduplicate to decisions
Multiple matching passages are grouped so you get one ranked card per decision, never duplicates.
Enrich with authoritative data
Official metadata — reference, date, outcome, business — comes from our database, never from the model.
Ground the summary
The overview is generated only from the retrieved decisions, cites references, and states its limitations.
Validate every citation
We check that each cited decision and passage really exists before showing you anything.
What the AI is not allowed to do
- ✕ Predict how your complaint would be decided
- ✕ Give legal advice or a recommendation
- ✕ Invent official metadata or outcomes
- ✕ Cite a decision it did not retrieve