365 PV Copilot
Enterprise AI for Pharmacovigilance
GVP Module VI · EMA 21 CFR Part 11 · FDA Human oversight

From a plain-text report to an audit-ready safety case.

An enterprise AI assistant that reads adverse-event reports in plain language, drafts a coded and enriched pharmacovigilance case, and submits it for medical review — inside Microsoft Teams. Specialists stay in control of every decision; every action is logged, hashed, and inspection-ready.

PV Case Assistant
Online · enterprise AI agent
MS TEAMS
65-year-old patient on warfarin developed severe nausea and dizziness after starting ibuprofen.
I've drafted a coded case for your review.
Case PV-0042DRAFT · awaiting review
ReactionNauseaDizziness
DrugsIbuprofenWarfarin
SeveritySevere  suggested
Signal⚠ Drug–drug interaction flagged
📚 Coded against MedDRA & WHODrug EditApprove ✓
Business impact

What it delivers to the safety organization

When AI absorbs the manual work of case intake and coding, specialists are freed for clinical judgement — and every case and AI decision is measured on a live dashboard, so the value is visible and accountable.

~50%

Faster case processing

Intake, coding and enrichment that took hours are drafted in minutes.

−70%

Manual data entry

The AI extracts and pre-fills case details, so specialists stop retyping reports.

~95%

Coding consistency

Standardized MedDRA & WHODrug suggestions reduce variation between reviewers.

100%

Audit trail coverage

Every action and AI decision captured, hashed and ready for inspection.

📊

Measured, not assumed. Case throughput, time saved, coding accuracy and audit completeness are tracked in a live Power BI dashboard, while every AI decision is monitored in Application Insights — giving leadership a clear, auditable view of ROI and compliance.


What it does

It turns reports into cases — you stay in control

Teams describe the event in plain language. The assistant does the heavy lifting and hands a clean, coded draft back for review.

📝

Reads free text

Extracts the adverse events, suspect drugs, patient details and dates from an unstructured report — no forms to fill in.

🏷️

Proposes the coding

Suggests the right MedDRA term for each reaction and the WHODrug entry for each product, ready for the reviewer to confirm.

🔬

Enriches & flags risk

Pulls in drug-safety context and flags potential drug–drug interactions, so important signals surface early.

📋

Manages the case

Creates, queries, updates, summarizes and deactivates cases across their full lifecycle — never deleting anything.

How it works

Every case passes through human approval

The assistant prepares the work; a qualified reviewer approves it. Nothing is written to the safety record until a person signs off.

Report described in Teams
Recognizes the usersigns them in & checks their role
Reads, codes & enriches the case
Extracts the eventsfrom free text
Proposes codingMedDRA & WHODrug
Flags interactionsand suggests severity
⏸ Reviewer confirms — human approval
Saved & loggedappend-only, tamper-evident audit trail
Built on Azure & Microsoft 365

It lives where your people already work

Same login, same security, no new application to learn — and your data stays inside your own Microsoft cloud tenant. Built entirely on enterprise tools you already own and trust.

CS

Copilot Studio

Where teams chat with the assistant in Teams.
API

FastAPI on Azure

The governed logic, controls and audit engine.
AI

Azure AI Foundry

The language model that reads and drafts cases.
DB

Azure SQL

Secure case records + append-only audit log.
ID

Entra ID

Role-based sign-in with existing work accounts.
TM

Teams

Where the safety team works every day.
AI

App Insights

Monitors every AI decision — full observability.
BI

Power BI

Live dashboard of safety KPIs and ROI.
The technical view

How it's built

It's a hybrid build. A low-code Copilot Studio agent gives the safety team a natural conversation in Teams and runs the human-approval step; it calls a pro-code FastAPI service on Azure that does the real work — reading text, proposing coding via the AI model, enriching, and enforcing the rules. The AI only ever suggests; all access control, writes, hashing and audit logging are deterministic and human-gated.

PV team
in Teams
Copilot Studio Low-code
Conversation topics
Entra ID sign-in
HITL confirm · YES / NO
Custom Connector
HTTPS
FastAPI on Azure Pro-code
Orchestrator + Case agent
RBAC · CRUD · audit
Calls AI · DDInter · OpenFDA
AI & data layer
Azure AI Foundry · LLM
Azure SQL · cases
Audit log + SHA-256
Build
Designed & iterated
UAT
Validated & signed off by the PV team
Live
Published to Teams
Governance & trust

Built for inspection, not just demos

Designed for regulated pharmacovigilance, with the controls and oversight quality, compliance and inspection teams expect.

GVP Module VI · EMA
ADR case lifecycle managed under qualified reviewer control.
21 CFR Part 11 · FDA
Electronic-signature approvals and a complete, append-only audit trail.
🧑‍⚖️

Human oversight (HITL)

Every write runs behind a human-in-the-loop approval — a qualified reviewer signs off before anything is recorded.

🔐

Roles & access controls

Entra ID with role-based access — PV Officer, Medical Reviewer, Auditor and Admin each see and do only what their role allows.

🏛️

Tamper-evident audit

An append-only log with SHA-256 before/after hashing — no record is ever hard-deleted, only deactivated or archived.

📡

Monitoring & observability

Every AI prompt, response and decision is captured in Application Insights and surfaced in Power BI.

📚

No invented codes. The assistant proposes terms grounded in the actual MedDRA and WHODrug dictionaries — it selects the correct official term rather than guessing, so every suggestion is traceable and verifiable.

More from the maker

More AI agents, built end to end

This is one of several real-world AI builds. See the full portfolio — and what could be built for your organization.

Visit lisekarimi.com