Dashboard

Regulatory review operations overview

Welcome to IND_Healthcare

Choose how you want to run this platform. You can switch modes at any time from Settings.

Demo Mode

No setup required

Explore the full MCP Orchestrator and all 7 specialist agents using realistic, pre-built sample submissions (New Drug Application, SAE narrative, Device Inspection report). Every agent returns illustrative, canned results instantly — no Anthropic API key, no configuration, and nothing leaves your browser.

  • Best for stakeholder walkthroughs and architecture demos
  • Works completely offline, instant responses
  • Synthetic data only — safe to show to anyone

Live Mode

Anthropic API key required

Run the same 7 agents on your own pasted text using real-time calls to the Anthropic Claude API (model claude-sonnet-4-6), made directly from this browser with your own key. No data or key is sent to any server operated by Angshuman AI Corp.

Your key is kept only in this browser's session storage and is cleared when the tab is closed. You can add or change it later from Settings.
IND_Healthcare — AI-Powered Regulatory Intelligence Platform. Concept/demo build for architecture and stakeholder walkthroughs.
Created by Angshuman Biswas
Submissions in Pipeline
38
▲ 6 new this week
Avg. Review Time Saved
63%
▲ vs manual baseline
PII / PHI Entities Masked
1,284
▲ DPDP-compliant
SAE Cases Flagged High
7
▼ awaiting reviewer sign-off

MCP Multi-Agent Pipeline — Live Status

Model Context Protocol orchestrates routing, context and policy across all specialist agents.
Click any agent node to open its workspace. Status reflects the most recent orchestrator run (Submission Intake → MCP Orchestrator).

Submission Volume by Type

Current pipeline composition (38 active cases)

SAE Severity Distribution

Classified Serious Adverse Event cases (last 30 days)

Recent Reviewer Activity

Latest actions from the CDSCO reviewer workbench
TimeCaseAgentActionReviewer
IND_Healthcare — AI-Powered Regulatory Intelligence Platform. Demo data is synthetic and for architecture illustration only; no real applicant or patient information is used.

New Submission Intake

Connectors to SUGAM (new drugs, clinical trials, imports) and MD Online (medical devices)

Quality Check & Dedup Pre-Screen

Step 1 of the ingestion pipeline (per architecture diagram)
1 Document / Data Ingestion
2 OCR / Text Extraction
3 Data Parsing & Structuring
4 Quality Check & Dedup Pre-Screen
In production, this layer would call OCR services and schema mappers from the Shared Services layer before handing structured JSON to the MCP Orchestrator. In this demo, pasted text is treated as already parsed.

Pre-Screen Result

Document structure recognised. No duplicate submission ID found in Data Layer. Ready for MCP routing.

MCP (Model Context Protocol) Layer — Secure Agent Communication Bus

Registry · Orchestrator · Context Memory · Security & Access Control · Audit Observability
MCP Registry
Tools & services catalogue for all 7 agents and shared services.
Context Memory
Session / case context shared across agents for a single submission.
Security & Access Control
RBAC, encryption, policy guard before any agent sees data.
Audit Observability
Immutable trace of every routing decision and agent call.

Run Full Pipeline on Selected Case

MCP routes the case sequentially through all 7 specialist agents, then hands a consolidated evidence pack to the reviewer.

Reviewer Queue — Human-in-the-Loop

CDSCO reviewers approve, query or reject AI recommendations. AI recommends; the reviewer decides.

Case Detail

Select a case from the queue to view its evidence pack.

Shared Services — MCP Tools & Enterprise Services

Common services consumed by every specialist agent via the MCP bus.

Data Layer

Raw Data Lake · Processed Data Repository · Anonymised Data Store · Knowledge Base · Model Store · Metadata & Audit DB

Model Layer

LLMs (Govt. / On-Prem / Secure) · Domain Models · ML Models · Model Training & Evaluation Pipeline

Governance, Security & Compliance

Applies across all layers — RBAC, encryption, DPDP/NDHM/ICMR compliance, audit, model governance, monitoring.

Audit Log — Immutable Trace

Every agent call, MCP routing decision and reviewer action is recorded for traceability (DPDP / ICMR compliance).
Audit entries are session-local in this demo. In production, the Audit Observability module within the MCP layer persists entries to the Metadata & Audit DB with model version, prompt/tool trace and reviewer identity.

Mode & API Configuration

Switch between Demo Mode (canned, illustrative agent outputs) and Live Mode (real-time calls to the Anthropic Claude API using your own key).
Live Mode notes
Live Mode sends agent prompts directly from this browser to https://api.anthropic.com/v1/messages using model claude-sonnet-4-6 with your API key in the x-api-key header and anthropic-dangerous-direct-browser-access: true. No data is sent to any server operated by Angshuman AI Corp. If the hosting domain triggers a browser CORS error, deploy behind a lightweight proxy or use Demo Mode for client walkthroughs. Always use synthetic / de-identified data for client demos until a production data-processing agreement and DPDP impact assessment are in place.

About IND_Healthcare

AI-Powered Regulatory Intelligence Platform — concept build

This application demonstrates an MCP-enabled multi-agent architecture for CDSCO regulatory review automation, covering data anonymisation (DPDP Act 2023, NDHM, ICMR), document summarisation, completeness & validation, SAE severity classification, duplicate detection, version comparison and draft report generation — with a human-in-the-loop reviewer workbench.

ArchitectureMCP Orchestrator + 7 specialist agents + Reviewer Workbench
PortalsSUGAM, MD Online (integration concept)
Compliance scopeDPDP Act 2023, NDHM, ICMR guidelines
Golden ruleAI recommends. CDSCO reviewer decides. System records why.
This is a concept/demo build for architecture and stakeholder walkthroughs. Production deployment requires legal, security and CDSCO subject-matter validation, integration with real SUGAM/MD Online APIs, and an approved data-processing agreement.