๐Ÿ›ก Responsible AI

NidaanPath AI is designed with transparency, safety, and human oversight at its core.

โŒ NidaanPath Does NOT

  • โœ—
    Diagnose any disease
  • โœ—
    Predict medical outcomes
  • โœ—
    Prescribe medication
  • โœ—
    Recommend specific medical tests
  • โœ—
    Change or stop medication
  • โœ—
    Replace healthcare professionals
  • โœ—
    Identify medical negligence
  • โœ—
    Guarantee medical safety
  • โœ—
    Decide which doctor is correct
  • โœ—
    Make autonomous clinical decisions

โœ… NidaanPath ONLY

  • โœ“
    Reconstructs the diagnostic journey timeline
  • โœ“
    Detects process gaps from confirmed records
  • โœ“
    Asks clarifying questions about process gaps
  • โœ“
    Matches new evidence to open gaps
  • โœ“
    Prepares evidence for safe clinician review
  • โœ“
    Displays source fragments for every finding
  • โœ“
    Requires patient confirmation before data enters journey
  • โœ“
    Uses deterministic rules for stagnation detection
  • โœ“
    Preserves and displays all clinical uncertainty
  • โœ“
    Generates evidence-grounded clinician packets

๐Ÿ›ก Three Mandatory Safety Gates

1

Patient Confirmation

No AI-extracted information enters the Diagnostic Journey Twin until the patient explicitly confirms or corrects it.

2

Deterministic Stagnation Rules

Journey state is determined by transparent Python rules only. Gemini cannot independently classify a journey as stagnant.

3

Clinician Escalation

Any medication uncertainty, contradiction or clinical ambiguity is escalated to the clinician packet. NidaanPath never resolves clinical uncertainty autonomously.

๐Ÿ”ง Technical Safeguards

๐Ÿงช
Synthetic Data Only

All demo records are synthetic. No real patient data is used anywhere.

๐Ÿ”‘
Server-Side API Only

Gemini API key is never exposed to the browser, JavaScript, or logs.

๐Ÿ‘
Source Traceability

Every extracted finding includes a source document and source fragment.

๐Ÿ“Š
Confidence Display

Extraction confidence is shown for every finding.

โœ…
Uncertainty Preservation

Uncertain fields are always preserved and marked, never silently resolved.

๐Ÿ“‹
Agent Audit Trail

Every agent tool call is logged with timestamp and safety-gate status.

๐Ÿ—‘
Temporary Processing

Uploaded files are processed server-side and are not shared.

๐Ÿท
Masked Identifiers

Patient identifiers use synthetic case codes only.

โš  Limitations

โš 
  • NidaanPath is a prototype MVP, not a regulated medical device.
  • Extraction accuracy depends on document legibility and format.
  • Mock AI mode uses deterministic responses, not real language understanding.
  • The system cannot verify document authenticity.
  • Journey reconstruction relies on confirmed documents only.
  • No Tamil OCR is supported in the current MVP.
โ† Back to Home