No legal conclusion without visible support

Veritas generates structured legal analysis where every claim is traceable to verified authority. Powered by GPT-4o and CourtListener's database of 18M+ real legal citations.

Why This Matters

The problem Veritas solves

LLMs hallucinate legal citations — studies show 6–35% fabrication rates in AI-generated legal content, with tools like ChatGPT citing cases that simply do not exist.

Lawyers require verifiable authority, not just answers. Every legal conclusion must trace back to real case law that an attorney can independently confirm.

Existing tools lack transparency into what is real vs. generated. Users cannot distinguish between a genuine citation and a confident hallucination.

Veritas introduces a verification layer that makes AI-generated legal reasoning auditable.

Pipeline Architecture

User InputIssue ExtractionQuery DecompositionCourtListener SearchHyDE Vector SearchCross-Encoder RerankingIRAC GenerationSelf-CritiqueDual VerificationLLM JudgeOutput

How Veritas Works

A multi-stage pipeline that retrieves, generates, verifies, and scores — ensuring trustworthy legal analysis.

1

Structured IRAC Analysis

Generates Issue, Rule, Application, Counterarguments, and Conclusion — grounded in retrieved case law.

2

Dual Citation Verification

Every citation is grounding-checked against the retrieval set and existence-checked via CourtListener's 18M+ database.

3

Agentic Self-Critique

The pipeline assesses retrieval quality, self-critiques analysis for logical gaps, and iterates to improve output.

4

LLM-as-Judge Scoring

Automated rubric-based evaluation scores reasoning quality, citation relevance, and conclusion support.

Confidence Rating System

Every citation receives a confidence rating based on dual verification

Strong
Retrieved from search AND verified in CourtListener
Moderate
Verified in CourtListener but not in retrieval set
Weak
In retrieval set but not found in CourtListener
Unverified
Neither retrieved nor verified

18M+

Legal Citations

in CourtListener database

13+

Quality Metrics

per evaluation run

4

Confidence Levels

per citation

2

Verification Layers

grounding + existence

Example Analysis

See what Veritas produces — a real IRAC analysis with verified citations

Input Question

“Can an employer terminate an employee for criticizing company leadership on social media?”

Issue

Whether an employer may lawfully terminate an at-will employee for posting criticism of company leadership on social media, considering the interplay of at-will employment doctrine and potential protections under the National Labor Relations Act.

Rule

Under the at-will employment doctrine, an employer may terminate an employee for any reason not prohibited by law. However, Section 7 of the NLRA protects employees' right to engage in “concerted activities” for mutual aid or protection. In Garcetti v. Ceballos, 547 U.S. 410 (2006), the Supreme Court held that public employee speech made pursuant to official duties is not protected by the First Amendment. The NLRB has extended protections to certain social media activity under Hispanics United of Buffalo, Inc., 359 NLRB 674 (2012).

Application

If the employee's social media post constitutes concerted activity — such as discussing workplace conditions with coworkers — the termination may violate the NLRA regardless of at-will status. Individual griping, however, is generally not protected.

Conclusion

The employer may terminate the employee under at-will doctrine unless the social media activity qualifies as protected concerted activity under the NLRA, in which case the termination would be an unfair labor practice.

Citation Verification

Garcetti v. Ceballos, 547 U.S. 410 (2006)
CL GRStrong
Hispanics United of Buffalo, Inc., 359 NLRB 674 (2012)
CL GRStrong
Pickering v. Board of Education, 391 U.S. 563 (1968)
CL Moderate

CL = Verified in CourtListener · GR = In retrieval set

Technology Stack

Production-grade infrastructure from retrieval to rendering

LLM & Embeddings

  • GPT-4o (reasoning)
  • GPT-4o-mini (planning)
  • text-embedding-3-small

Retrieval & Storage

  • CourtListener API v4.3
  • Qdrant vector DB
  • PostgreSQL + Redis

Application Layer

  • FastAPI + SSE streaming
  • Next.js App Router
  • Tailwind CSS + shadcn/ui

Explore Veritas

Jump to any section of the application