Fix the reliability gap.
General-purpose AI models hallucinate legal citations, fabricate case law, and produce unreliable advice. When an attorney submits a brief with fake citations, courts sanction them. When a self-represented litigant follows bad AI advice, they lose their case.
The reliability gap in legal AI is not a minor inconvenience — it is a liability risk that undermines trust in the entire system.
A Retrieval-Augmented Generation (RAG) engine purpose-built for legal applications. Every response is grounded in verified legal data. Every citation is validated against its source. Every answer carries a confidence score. Every query leaves a complete audit trail.
This is not a chatbot with a legal skin. It is infrastructure for trustworthy legal AI.
graph LR
Q[User Query] --> EMB[Embedding]
EMB --> VS[Vector Search]
VS --> CA[Context Assembly]
CA --> LLM[LLM Generation]
LLM --> CV[Citation Validator]
CV --> CS[Confidence Scorer]
CS --> AL[Audit Logger]
AL --> R[Verified Response]
subgraph RAG Pipeline
EMB
VS
CA
end
subgraph Validation Layer
CV
CS
AL
end
| Audience | How This Helps |
|---|---|
| Legal aid organizations | Reliable AI assistance without hallucination risk |
| Pro bono attorneys | Research tool they can trust for real case work |
| Court self-help centers | Provide accurate, sourced information to litigants |
| Justice tech platforms | Embed vetted AI into their applications |
- RAG pipeline with verified legal corpus
- Citation validation — every citation checked against source material
- Confidence scoring per response (high / medium / low / insufficient)
- Complete audit trail — query, sources, generation, validation logged
- Jurisdiction-aware retrieval — filter by state, federal, or topic
- Source transparency — full provenance chain for every answer
- Pluggable LLM backend (OpenAI, Anthropic, local models)
- FastAPI-based REST interface
| Layer | Technology |
|---|---|
| Language | Python 3.11+ |
| RAG Framework | LangChain |
| Vector Store | ChromaDB |
| API | FastAPI |
| Validation | Pydantic |
| Testing | pytest |
| Linting | Ruff + mypy |
git clone https://github.com/dougdevitre/vetted-legal-ai.git
cd vetted-legal-ai
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
uvicorn src.vetted_legal_ai.api.routes:app --reload
# API is now running at http://localhost:8000 — see /docs for Swagger UIimport httpx
response = httpx.post("http://localhost:8000/query", json={
"question": "What are the grounds for eviction in Missouri?",
"jurisdiction": "MO",
"max_sources": 5,
})
result = response.json()
print(result["answer"])
print(f"Confidence: {result['confidence']}")
for cite in result["citations"]:
print(f" [{cite['status']}] {cite['case_name']} — {cite['citation']}")See examples/basic_query.py for a complete working example.
| Feature | Status |
|---|---|
| RAG pipeline with ChromaDB vector store | In Progress |
| Citation validation against source corpus | In Progress |
| Confidence scoring per response | Planned |
| Jurisdiction-aware retrieval filtering | Planned |
| Pluggable LLM backend (OpenAI, Anthropic, local) | Planned |
| Complete audit trail with query provenance | Planned |
This repository is part of the Justice OS open-source ecosystem — 32 interconnected projects building the infrastructure for accessible justice technology.
| Repository | Description |
|---|---|
| justice-os | Core modular platform — the foundation |
| justice-api-gateway | Interoperability layer for courts |
| legal-identity-layer | Universal legal identity and auth |
| case-continuity-engine | Never lose case history across systems |
| offline-justice-sync | Works without internet — local-first sync |
| Repository | Description |
|---|---|
| justice-navigator | Google Maps for legal problems |
| mobile-court-access | Mobile-first court access kit |
| cognitive-load-ui | Design system for stressed users |
| multilingual-justice | Real-time legal translation |
| voice-legal-interface | Justice without reading or typing |
| legal-plain-language | Turn legalese into human language |
| Repository | Description |
|---|---|
| vetted-legal-ai | RAG engine with citation validation |
| justice-knowledge-graph | Open data layer for laws and procedures |
| legal-ai-guardrails | AI safety SDK for justice use |
| emotional-intelligence-ai | Reduce conflict, improve outcomes |
| ai-reasoning-engine | Show your work for AI decisions |
| Repository | Description |
|---|---|
| evidence-vault | Privacy-first secure evidence storage |
| court-notification-engine | Smart deadline and hearing alerts |
| justice-analytics | Bias detection and disparity dashboards |
| evidence-timeline | Evidence timeline builder |
| Repository | Description |
|---|---|
| court-doc-engine | TurboTax for legal filings |
| justice-workflow-engine | Zapier for legal processes |
| pro-se-toolkit | Self-represented litigant tools |
| justice-score-engine | Access-to-justice measurement |
| justice-app-generator | No-code builder for justice tools |
| Repository | Description |
|---|---|
| justice-persona-simulator | Test products against real human realities |
| justice-experiment-lab | A/B testing for justice outcomes |
| Repository | Description |
|---|---|
| digital-literacy-sim | Digital literacy simulator |
| legal-resource-discovery | Find the right help instantly |
| court-simulation-sandbox | Practice before the real thing |
| justice-components | Reusable component library |
| justice-dev-starter-kit | Ultimate boilerplate for justice tech builders |
Built with purpose. Open by design. Justice for all.
This project is provided for informational and educational purposes only and does not constitute legal advice, legal representation, or an attorney-client relationship. No warranty is made regarding accuracy, completeness, or fitness for any particular legal matter. Always consult a licensed attorney in your jurisdiction before making legal decisions. Use of this software does not create any professional-client relationship.
I build AI-powered platforms that solve real problems. I also speak about it.
CoTrackPro · admin@cotrackpro.com
→ Hire me: AI platform development · Strategic consulting · Keynote speaking
AWS AI/Cloud/Dev Certified · UX Certified (NNg) · Certified Speaking Professional (NSA) Author of Screen to Screen Selling (McGraw Hill) · 100,000+ professionals trained