Grounded answers, or none at all.

• Code decides, not the model

  Plain, testable checks decide what’s shown, with or without AI.

• Every claim is cited

  Each sentence points to the source passage it came from.

• Every request is recorded

  A full trace is saved, whether it answered or refused.

• Engineers

  Start from guardrails that already work, with tests.

• Clinical safety teams

  See refusal, citation and audit enforced in code.

• Researchers and educators

  An offline testbed with 2,008 test questions.

• Reviewers and auditors

  Trace any decision to its check and its source.

Not for patient care. GroundCheck isn’t clinically validated or cleared by any regulator.

1. Check the question

   1. 1PII redactionHides emails and long numbers
   2. 2Scope checkBlocks prompt injection
   3. 3Rate limitStops one caller using it all

2. Find the evidence

   1. 4RetrieveFinds the closest passages
   2. 5Retrieval gateRefuses weak matches
   3. 6Source coverageRefuses terms no source mentions

3. Draft and verify

   1. 7GenerateDrafts claims with citations
   2. 8Schema validationRejects malformed output
   3. 9Grounding checkChecks each claim against its source
   4. 10Dosage guardVerifies every dose

4. Decide and record

   1. 11Decision and auditAnswers or refuses, and saves why

• Available

  Extractive

  Answers are built from sentences in your sources.

  • No API key
  • Fully offline
• Available

  Cloud model

  Any OpenAI-compatible model drafts the answer. Groq is the default.

  • Needs an API key
  • Sends redacted question and passages
• Planned

  Local AI model

  An open-source model on your own machine drafts the answer.

  • No API key
  • Offline after one download

If a model fails or times out, GroundCheck falls back to the extractive answer.

Runs offline in extractive mode, so these results measure the checks, not a language model.

• Word-based coverage checkCan refuse over everyday words like “year” or “email”.
• Broad extractive draftsCan include related passages the question didn’t ask about.
• Basic PII redactionEmails and long numbers only. Not de-identification.
• Synthetic tuningThresholds need re-tuning for real documents.
• Self-referential testsAnswerable cases come from the corpus itself.
• Local audit logSeveral instances would need a shared store.
• No review queue yet“Routed for review” is a record, not a workflow.

• Available now

  Grounded retrieval core

  Prove that a clinical assistant can refuse, cite and record by design.

  What’s included

  • Eleven-stage pipeline with deterministic guards
  • Citations, trace and audit record for every request
  • 2,008-case evaluation that gates the build
  • Tuning dashboard and JSON API
• Planned

  Local AI models

  AI-written answers that never leave the machine.

  What’s included

  • Hardware check for memory, GPU and Apple silicon
  • Open-source models recommended to fit that hardware
  • One-click download and loading through a local runtime such as Ollama
  • Automatic fallback to extractive answers when no model is available
• Planned

  Accounts and data foundation

  Give every user, source and decision a durable, queryable home.

  What’s included

  • Relational database: SQLite for a single machine, PostgreSQL or MySQL in production, with versioned migrations
  • Pluggable vector database: FAISS locally, pgvector or Qdrant at scale
  • Sign-in with securely hashed passwords, multi-factor authentication and session management
  • Clinician, reviewer and admin roles, with every audit record tied to a user
• Planned

  Your organisation’s knowledge

  Answer from a hospital’s own approved guidelines, formularies and protocols.

  What’s included

  • Import PDF, Word and HTML, with section-aware chunking
  • Source ownership, versions, effective dates and expiry
  • Approval workflow before content can be cited
  • Automatic re-indexing, and evaluation cases generated from each source
• Planned

  Clinician review and governance

  Turn every refusal and flagged answer into a reviewed, closed case.

  What’s included

  • Review queue with assignment, escalation and service-level timers
  • Clinician feedback that becomes new sources or evaluation cases
  • Clinical safety case and hazard log for each release
  • Usage, refusal and override reports for governance committees
• Planned

  Security, privacy and access

  Meet hospital IT and information governance requirements.

  What’s included

  • Single sign-on with SAML and OpenID Connect, on top of the built-in accounts
  • Full PHI de-identification, beyond pattern matching
  • Encryption in transit and at rest, and tamper-evident audit logs
  • On-premises and private cloud deployment, with retention controls
• Planned

  Patient-aware safety checks

  Check answers against the patient in front of the clinician, not just the literature.

  What’s included

  • Allergy, contraindication and drug-interaction checks
  • Dose ranges adjusted for weight, age, and kidney and liver function
  • Explicit handling of paediatrics, pregnancy and high-alert medicines
  • Refusal when required patient data is missing
• Planned

  EHR integration

  Bring grounded answers into the clinical workflow, where decisions are made.

  What’s included

  • HL7 FHIR R4 access to medications, allergies, labs and problems
  • SMART on FHIR launch from inside the electronic health record
  • CDS Hooks cards with citations at the point of ordering
  • Write-back of reviewed notes, with consent and audit
• Planned

  CT and MRI imaging

  Apply the same principle to scans: show the evidence, or abstain.

  What’s included

  • DICOM and DICOMweb intake from PACS, with de-identification
  • Pluggable models for CT and MRI tasks, such as segmentation and triage
  • Abstention on unfamiliar scanners, protocols or low-confidence results
  • Findings tied to image regions, a radiologist sign-off step, and DICOM Structured Reports
• Planned

  Model training studio

  Train imaging models on your own data and hardware, then reuse them anywhere in the platform.

  What’s included

  • Pick a folder of images or a registered dataset, with format and label checks
  • Detects NVIDIA and Apple silicon GPUs and lets you choose one, falling back to the CPU
  • Live progress, validation metrics and the option to stop and resume
  • A versioned model library, ready to use in the imaging pipeline
• Planned

  Production operations

  Run across many sites, and catch problems before clinicians do.

  What’s included

  • Multi-site tenancy, high availability and Kubernetes deployment
  • Monitoring of refusal rates, latency and data drift, with alerts
  • Versioned models and sources, staged rollout and one-step rollback
  • Incident reporting and post-market surveillance

Want to shape what comes first, or help build it? Join the discussion on GitHub.

git clone https://github.com/sumitgundawar/GroundCheck.git
cd GroundCheck
python3.12 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python scripts/build_index.py
uvicorn app.main:app --port 8000

git clone https://github.com/sumitgundawar/GroundCheck.git
cd GroundCheck
docker build -t groundcheck .
docker run -p 7860:7860 groundcheck

Full setup, configuration and Windows instructions are in the README.