Ground responses in evidence
Lexical retrieval helps locate specific terms and identifiers, while semantic search finds related meaning. Together, they supply relevant source context to the LLM, reducing reliance on pretrained knowledge alone.
Enterprise GraphRAG · Evaluated on CINECA Leonardo
Make enterprise knowledge usable by every AI agent.
Verbis Graph transforms complex document collections into connected, source-grounded context - so AI systems can retrieve what matters, trace relationships, and show the evidence behind an answer.
- Hybrid graph + vector retrieval
- Source-linked answers
- Private cloud deployment
- No customer-data model training
- Ontology aware
Evidence path
Source grounded
“Which supplier obligations changed after Policy 24-B?”
04 supporting sources
03 linked relationships
01 traceable answer
Illustrative product view
Evaluated and available across your enterprise ecosystem
CINECA Leonardo
Microsoft Marketplace
AWS Marketplace
Google Cloud Marketplace
Product walkthrough
See how knowledge becomes easier to work with.
Follow the Verbis Graph workflow from workspace setup and document upload to graph exploration and grounded chat.
- 01 Create a user and workspace
- 02 Upload and organize documents
- 03 Explore the knowledge and ontology graph
- 04 Check answer traceability
Watch the product workflow, from the first workspace to evidence-led questions.
Why Verbis Graph
AI cannot reason reliably over knowledge it cannot connect.
Enterprise information is scattered across contracts, policies, reports, research, and operational documents. Standard semantic search can find similar text. Verbis Graph is built to uncover how the facts relate - and preserve the path back to source.
Knowledge is fragmented
Relevant facts live across files, formats, and teams.
Connected retrieval
Entities and relationships bring cross-document context into one navigable graph.
Find the full answer
Investigate beyond a single paragraph or isolated similarity match.
Answers are difficult to verify
Opaque summaries create risk for high-value decisions.
Evidence paths
Citations and source context let teams inspect how an answer was formed.
Review with confidence
Give analysts, auditors, and domain experts a traceable basis for action.
Agent context is expensive
Sending large amounts of irrelevant text increases cost and noise.
Scoped hybrid retrieval
Vector, lexical, and graph signals narrow context to relevant evidence.
Use context deliberately
Design agent workflows around smaller, better-grounded evidence sets.
Reducing hallucination risk & ambiguity
Grounded answers.
Less ambiguity.
Verbis Graph helps reduce hallucination risk and ambiguity by grounding LLM responses in relevant enterprise knowledge.
Lexical, semantic, and knowledge-graph retrieval combine precise terminology, conceptual meaning, entities, relationships, and supporting evidence across documents.
Clarify meaning and relationships
Graph context connects entities and relationships across documents, helping clarify how facts relate. Ontology-aware context can add domain-specific terminology and structure where configured.
Make answers verifiable
Source-linked citations and retrieved evidence let teams inspect supporting passages and check the basis of a response, supporting review in legal, compliance, financial, and other knowledge-intensive workflows.
Evaluate the evidence. Validate the outcome.
Our CINECA Leonardo benchmarks measure retrieval quality. In a focused PoC, evaluate answer grounding, citation accuracy, and answer correctness using your own documents and chosen LLM.
One knowledge layer, multiple AI experiences
Index once. Serve grounded context wherever teams work.
Verbis Graph creates a reusable knowledge layer between enterprise content and your LLM or agent stack. Teams can search, inspect the graph, or retrieve context through an API without rebuilding knowledge for every interface.
Discuss your architectureIngest
PDF, DOCX, TXT, CSV, OCR images, and transcripts.
Connect
Extract entities and relationships; combine lexical, vector, and graph retrieval.
Retrieve
Return multi-hop context, citations, and evidence paths for complex questions.
Integrate
Support assistants, AI agents, research workflows, and enterprise applications through API access.
Built for knowledge-intensive work
Legal & compliance
Financial research
Corporate knowledge
Scientific evidence
Vendor & contract analysis
HPC evaluation
Measured across biomedical, scientific, financial, and legal retrieval.
CINECA Leonardo
Evaluated on high-performance infrastructure, including local 32B-model testing on 4× NVIDIA A100 GPUs.
Biomedical · NFCorpus
0.592
MRR
Latest embedding configuration, following an initial 0.303 baseline.
Scientific · SciFact
0.721
MRR
Evidence-retrieval evaluation across 300 benchmark queries.
Financial · FiQA
95.3%
Any-rank coverage
With 0.584 MRR and 85.3% mean recall.
TIME TO FIND RELEVANT EVIDENCE
≈ 33 sec
Per Query
Across ~57,600 documents in the FiQA benchmark.
Dataset-specific results are not directly comparable. Metrics reflect controlled benchmark configurations and the stated LegalBench-RAG checkpoint; they are evidence for technical evaluation, not a guarantee of production performance.
Full evaluation report
View Full Evaluation Report
Review the complete Verbis Graph evaluation on HPC infrastructure, including the benchmark methodology, retrieval configurations, datasets, measured results, and technical analysis.
Your deployment boundary
Verbis Graph
Knowledge graph
Vector index
Retrieval API
Graph explorer
Deployment & security
Fit Verbis Graph into your architecture-not the other way around.
Evaluate through a hosted demonstration or deploy the container into a controlled cloud environment. The platform’s layered Python/FastAPI architecture supports API integration and enterprise technical review.
- 01Cloud-aligned procurement
Buy through established marketplace channels and existing vendor processes.
- 02Container-based deployment
Plan a VPC-oriented architecture for document-sensitive use cases.
- 03Integration-ready API
Connect retrieval to your agent, application, or LLM orchestration layer.
Focused enterprise evaluation
Prove value on your documents in one month.
A focused PoC turns one priority knowledge problem into a measurable decision: proceed, refine, or stop-with evidence your technical and business teams can review.
Plan a 1-Month PoC- Week 1Define the use case
Agree questions, corpus, deployment boundary, and success criteria.
- Week 2Index and connect
Build the graph and configure retrieval for your documents.
- Week 3Test real workflows
Evaluate relevance, evidence quality, usability, and latency.
- Week 4Review the decision
Document outcomes and define the path to production.
Enterprise procurement
Evaluate and procure through the cloud you already use.
Move from technical evaluation to purchasing through familiar cloud marketplace routes, aligned with enterprise billing, vendor review, and deployment planning.
Available
AWS Marketplace
Verbis Graph has successfully passed the AWS FTR, confirming our architecture meets AWS best‑practice standards. .
AWS MarketplaceAvailable
Microsoft Marketplace
Evaluate through Azure Marketplace or AppSource for Microsoft-oriented cloud teams.
Microsoft MarketplaceAvailable
Google Cloud Marketplace
Marketplace availability is in preparation for teams operating on Google Cloud.
Google Cloud Marketplace
Prodigy AI Solutions is a member of NVIDIA Inception.
Supporting our roadmap for scalable high-performance enterprise
retrieval, GraphRAG, and AI-agent infrastructure.
Published in the NVIDIA App Catalog
Verbis Graph is listed in NVIDIA’s enterprise Applications catalog, alongside DPU- and GPU-accelerated applications, tools, and services built on NVIDIA platforms for AI, cloud computing, data science, and more.
View Verbis Graph in the NVIDIA App CatalogExplore with AI
Ask about Verbis Graph in your AI workspace.
Use a dedicated Verbis Graph AI assistant to explore solution architecture, deployment options, marketplace routes, integrations, and PoC planning.
ChatGPT
Explore architecture, deployment, integration, marketplace options, and focused PoC design with the Verbis Graph Enterprise Solution Architect.
Gemini
Explore Verbis Graph capabilities, deployment considerations, integration approaches, and PoC planning in Gemini.
Note: Responses are informational. Specific enterprise architecture, security, commercial, and deployment requirements should be confirmed directly with Prodigy AI Solutions.
Frequently asked questions
What enterprise teams ask before a PoC.
Review common questions around evaluation, deployment, integration, security, procurement, and what happens after a focused 1-month Verbis Graph PoC.
Can we evaluate Verbis Graph with our own documents and use cases?
Yes. A focused PoC can be structured around representative enterprise documents, real questions, target workflows, and agreed success criteria so your team can evaluate relevance and fit in a realistic environment.
How does Verbis Graph provide traceable, source-grounded answers?
Verbis Graph combines graph-based context with semantic/vector and lexical retrieval so retrieved information can be connected back to supporting sources and inspected as part of the answer workflow.
Where does our data run, and what deployment options are available?
Deployment can be aligned with your cloud, security, residency, and operational requirements. The appropriate architecture is defined according to the selected deployment path and enterprise constraints.
Can Verbis Graph work with our existing LLMs and AI stack?
Yes. Verbis Graph can be introduced as part of a broader enterprise AI architecture and can complement existing LLMs, applications, agents, APIs, and retrieval workflows where supported.
How are document permissions and access controls handled?
Access-control requirements should be incorporated into the deployment architecture and PoC scope so retrieval respects the authorization model defined for the enterprise environment.
How do we measure whether the 1-month PoC is successful?
Success criteria can be agreed before the evaluation and may include retrieval quality, source traceability, answer usefulness, review time, latency, deployment fit, and workflow value using representative enterprise content.
Can we procure Verbis Graph through our existing cloud marketplace?
Verbis Graph is available through AWS and Microsoft marketplace channels. Google Cloud Marketplace availability is coming soon.
Does Verbis Graph replace our existing search, vector database, or AI platform?
Not necessarily. Verbis Graph can complement existing document systems, search components, vector retrieval, LLMs, applications, and AI-agent workflows instead of requiring a full replacement.
What happens after the 1-month PoC?
The PoC is intended to support a clear decision: proceed toward a broader deployment, refine the use case or architecture, or stop without unnecessary expansion.
How can hallucination-risk reduction be evaluated?
Retrieval metrics such as MRR, Recall, Coverage, and Hit@K measure how effectively relevant evidence is retrieved, but they do not directly measure hallucinations in generated answers. Downstream impact can be evaluated using a controlled question set derived from the customer's knowledge base. The same LLM is tested with and without Verbis Graph context: answer correctness and unsupported claims can be compared across both configurations, while the Verbis-enabled responses can additionally be evaluated for factual grounding and citation accuracy against the source documents.
Have a deployment, security, or evaluation question specific to your environment?
Plan a 1-Month PoC
Ready to test connected retrieval?
Plan a 1-Month PoC