Skip to main content
Advanced AI Technology

Find Hidden Contradictions Keyword Search Misses

Ask "Which experts contradict each other about causation?" and get answers cited to the page. LexHound's Graph RAG understands relationships between people, claims, and facts across all your documents—not just keyword matches.

The Problem with Keyword Search

Traditional search tools use keyword search: You search for "causation" and get every page mentioning "causation." Then you manually read 200 pages to find if any witnesses contradict each other.

This is slow and incomplete. You can't ask complex relational questions like:

  • "Which two expert witnesses contradict each other about the timing of the surgery?"
  • "Find all mentions of back pain between January and March 2024"
  • "Did any witness change their testimony between deposition and trial?"
  • "Which doctors disagree about the standard of care?"

Keyword search can't answer these because it doesn't understand relationships between entities (people, dates, events, claims).

How Graph RAG Solves This

Graph RAG (Retrieval-Augmented Generation with Knowledge Graphs) is a new AI technology that understands who said what, when, to whom, and how it connects to other facts.

When you upload documents to LexHound, our AI:

  1. Extracts entities: People (witnesses, doctors, plaintiffs), dates, locations, medical conditions, treatments, claims
  2. Maps relationships: "Dr. Smith testified about causation on March 15, 2024" → "Dr. Jones testified about causation on April 3, 2024" → "Their opinions contradict"
  3. Builds a knowledge graph: A map of how all entities connect across all documents
  4. Answers relational queries: You ask complex questions, AI searches the graph (not just keywords) and finds connected facts

Example: Finding a Contradiction in a Surgical Error Case

Example workflow, not a client matter.

Medical Malpractice Case: Surgical Error

Documents:

  • 12,000 pages of medical records
  • 6 expert depositions (3,000 pages)
  • Operative reports, nursing notes, imaging studies

Attorney Query (using Graph RAG):

"Which expert witnesses contradict each other about the timing of when the surgeon was notified about post-operative complications?"

LexHound Answer:

Found 2 contradictions:

Dr. Sarah Mitchell (Defense Expert):

"The surgeon was notified immediately at 2:47 PM via phone call from the charge nurse."
(Source: Mitchell Deposition, Page 234, Lines 15-18)

Charge Nurse Emily Rodriguez (Fact Witness):

"I tried calling Dr. Johnson at 2:47 PM but got no answer. I left a voicemail. He called back at 4:15 PM."
(Source: Rodriguez Deposition, Page 456, Lines 8-12)

How Long Would This Take Manually?

  • Read 6 expert depositions: 8-12 hours
  • Cross-reference nursing notes with deposition testimony: 4-6 hours
  • Create timeline of phone calls manually: 2-3 hours
  • Total: roughly 14-21 hours of reading and cross-referencing

Graph RAG vs Keyword Search: Side-by-Side Comparison

Question TypeKeyword SearchGraph RAG (LexHound)
"Find all mentions of back pain" Works
Returns 200 pages with "back pain"
Works Better
Returns results + timeline of when pain mentioned
"Which two experts contradict each other?" Doesn't Work
Can't understand relationships between witnesses
Works
Compares all expert opinions automatically
"Find evidence between Jan-Mar 2024" Partial
Matches dates as text; misses relative dates like "post-op day 3"
Works
Filters by date range automatically
"Who was present during the surgery?" Partial
Returns mentions but you manually connect people to event
Works
Lists all people connected to "surgery" event
"Did any witness change their story?" Doesn't Work
Can't compare same person across multiple documents
Works
Compares deposition vs trial testimony automatically

Specific Use Cases for Attorneys

Medical Malpractice Cases

  • Expert Witness Contradictions: "Which experts disagree about standard of care?"
  • Timeline Gaps: "What happened between the surgery and when the surgeon was notified?"
  • Medical Causation: "Which doctors link the surgery to the patient's current condition?"
  • Standard of Care Deviations: "Compare the nursing notes with the hospital protocol we uploaded and list any deviations"

Personal Injury Cases

  • Witness Inconsistencies: "Did the defendant's story change between police report and deposition?"
  • Pre-Existing Conditions: "Find all mentions of back pain before the accident date"
  • Multiple Defendants: "Which defendant blamed the other for the accident?"
  • Damages Timeline: "When did the plaintiff first report each injury symptom?"

Product Liability Cases

  • Pattern Identification: "Which plaintiffs experienced the same side effect?"
  • Temporal Relationships: "When did each plaintiff start using the product relative to symptom onset?"
  • Defendant Knowledge: "Did the manufacturer know about this defect before marketing?"

How LexHound Implements Graph RAG

LexHound combines three types of search for comprehensive results:

1. Vector Search

Understands semantic meaning. "Surgical error" matches "operative mistake" even though words differ.

2. Keyword Search

Traditional exact match. Ensures you don't miss specific terms like case citations or medical codes.

3. Graph Search

Finds entity relationships. "Which experts contradict?" searches the knowledge graph, not keywords.

This triple retrieval approach means relevant material surfaces whether it's an exact keyword match, a semantically similar passage, or connected through entity relationships.

Frequently Asked Questions

What is Graph RAG in simple terms?

Graph RAG is AI technology that understands relationships between people, facts, and events in your documents. Instead of just finding keywords, it understands "who said what to whom, when, and how it connects to other facts." This lets you ask complex questions like "Which witnesses contradict each other?" and get answers with page citations.

How is this different from ChatGPT or Claude?

General AI chat tools work from what you paste or upload into one conversation. They don't build a map of every person, date, and claim across your whole case file, so cross-document questions depend on what you happened to give them.

How is this different from keyword search in other legal tools?

Keyword search finds the word; it doesn't connect who said what, when. With keyword search you search for each expert separately and compare their testimony yourself. Graph RAG maps the relationships, so you can ask "Which experts contradict each other?" directly.

Do I need to understand how Graph RAG works to use it?

No. You just ask questions in plain English. LexHound handles all the technical complexity behind the scenes. You don't need to know about "knowledge graphs" or "entity extraction"—just ask your question like you'd ask a paralegal.

How accurate is Graph RAG at finding contradictions?

Graph RAG is highly accurate at identifying potential contradictions, but you should always verify. LexHound shows you the exact source (document, page, and line number for transcripts, with the passage highlighted) for each piece of evidence, so you can verify before using it in court. Think of it as a supercharged paralegal: it finds leads, you verify them.

Can Graph RAG handle handwritten notes or poor-quality scans?

Graph RAG works on the text extracted from documents. LexHound reads scanned and faxed records with OCR (Optical Character Recognition). Typed text reads reliably; handwriting is harder, so spot-check messy handwritten notes against the cited page.

How many documents can Graph RAG analyze at once?

Unlimited. LexHound has handled case files of more than 200,000 pages and mapped entity relationships across the entire set. The more documents you upload, the more connections Graph RAG can find.

Does Graph RAG work in languages other than English?

Yes. LexHound reads documents in any language, and Graph RAG maps relationships across them.

How long does it take to build the knowledge graph?

Graph RAG builds the knowledge graph in the background as your documents process, and we email you when it's done.

Can I export what Graph RAG finds?

Yes. Export findings, with their citations, to PDF or Word (DOCX) for your case file, deposition prep, or expert review.

Stop Missing Critical Contradictions Hidden in Your Documents

Graph RAG finds entity relationships and contradictions keyword search misses. Try it on your next case.