AI Assistant for Pharmacists

For our Master’s HCI project, we explored how AI can support hospital pharmacists without replacing their expertise. We designed a decision-support tool that helps them work more efficiently and confidently in fast-paced clinical environments.

The Objective:

To design an AI-assisted decision-support system that enhances pharmacists’ accuracy and efficiency by analyzing medication data, identifying potential risks, and providing actionable insights to support safe prescription dispensing.

My Contribution:

Research, Interviews, Ideation, Analysis, Wireframes & Prototype, UI-UX

Team:

4 Student Designers

Timeline:

October 2025 - November 2025

What did I do:

01

Synthesized secondary research, 2 in-depth pharmacist interviews, and competitor analysis to identify medication errors as the core area

02

Mapped complex clinical workflows and pharmacist mental models, breaking down decision-making across various parameters

03

Defined and strategically scoped a high-stakes AI intervention by evaluating three potential domains and narrowing the focus

04

Designed a seamless AI plugin for existing pharmacy systems that minimized learning curves and eliminated the need for redesign.

05

Developed and prototyped multi-scenario AI interaction flows to enhance prescription accuracy and clinical confidence

06

Conducted expert evaluation and iterative refinement with a practicing hospital pharmacist, improving AI transparency and workflow clarity

Design Process ✦

Project Brief

For our Master’s Introduction to HCI project, we were given an open-ended challenge to design an AI solution that supports professionals without replacing them. In a time when AI is often seen as a threat to jobs, this project explored how intelligent systems can empower workers, augmenting their expertise, enhancing decision-making, and preserving human authority rather than substituting it.

Why Pharmacists

At the outset, we chose to explore hospitals because they are high-stakes, multidisciplinary environments where small errors can have serious consequences. After exploring multiple roles within the hospital ecosystem, we chose to focus on clinical pharmacists because they sit at a critical decision-making junction between doctors, nurses, and patients. While doctors prescribe and nurses often detect errors, pharmacists are uniquely positioned to validate prescriptions, review medical histories, and identify drug–drug interactions before medications reach patients.

Unlike doctors, whose workload makes additional systems burdensome, or nurses, who are constantly mobile, pharmacists operate within structured digital workflows, making them ideal candidates for an assistive AI intervention. Their authority to modify prescriptions, combined with their responsibility for accuracy and patient safety, made them the most impactful and practical users for our solution.

6.5/ 100

Patients are administered incorrect medication

30-70%

Nurses and pharmacists identify medication errors

1.5 Million

People are affected every year due to medication errors

Primary Research

After reviewing existing literature, we conducted user interviews to uncover pharmacists’ real-world challenges and motivations. This helped us identify clear opportunities where AI could meaningfully assist without replacing their expertise.

Synthesizing Research:

Insights from Research

Revealed recurring challenges such as inventory strain, prescription verification pressure, and the cognitive load of preventing medication errors.

Affinity Mapping

Synthesized qualitative data using affinity mapping to identify recurring themes, patterns, and high-impact opportunity areas.

Mental Model

Analyzed pharmacists’ mental models to understand their decision-making processes and determine where AI could be integrated.

Interview Insights:

Inventory management and stock reconciliation were major friction points, requiring constant manual tallying between physical and system records.

Dispensed drug verification was cognitively demanding, especially when colleagues were unavailable for double-checking during peak hours.

Auditing processes were repetitive yet high-risk, involving cross-checking sold drugs against system stock with zero room for error.

Prescription validation required intense scrutiny, including reviewing diagnosis alignment, dosage accuracy, strength, and potential drug interactions.

AI in Use:

Unlike traditional rule-based systems that rely on fixed “if–then” logic and predefined databases, our solution leverages assistive AI powered by predictive analytics to analyze diverse data sources such as EHRs, clinical literature, drug metadata, and patient history, to detect both known and previously unseen drug interaction risks.

By learning from structured and controlled unstructured data, the system adapts to evolving medical knowledge, provides broader coverage, and enables context-sensitive recommendations. Rather than replacing human judgment, the AI acts as a smart second pair of eyes that flags inconsistencies and provides clear, evidence-backed insights so pharmacists can make faster, safer dispensing decisions.

HIPAA Compliance Cosiderations:

Designed the system to be HIPAA-compliant, ensuring all patient data remains on-premise within the hospital network.
Implemented the minimum necessary data principle, with AI trained only on de-identified or internally approved clinical data.
Secured access through role-based authentication and detailed audit logs to ensure accountability and traceability.
Ensured all AI recommendations are explainable, transparent, and fully reviewable by pharmacists, preserving human authority.
Leveraged federated learning for cross-location model improvements without sharing raw patient data, maintaining privacy across institutions.

Ideation

After synthesizing research and interviews, we created early sketches to explore how AI could assist pharmacists in detecting drug anomalies within their existing workflows.

Proposed User Journey & Scope

We translated our research into a focused user journey that supports prescription validation through multi-source data integration. The solution was designed as assistive (not replacive), limited to clinical pharmacists and outpatients, excluded controlled substances for compliance, and built as a plugin to integrate seamlessly into existing pharmacy systems.

Final Solution

Workflow Initiation: The workflow begins when pharmacists scan or manually enter the Patient ID or Prescription ID, which activates the AI system to start analyzing and verifying the prescription.

Unified Clinical Overview: The interface presents key patient information such as allergies, ongoing conditions, current medications, and essential demographics alongside prescription and prescriber details, bringing all critical clinical context into one centralized dashboard for informed decision making.

Intelligent Conflict Detection: The AI analyzes prescriptions for potential risks including drug allergy conflicts, drug drug interactions, contraindications, and diagnosis mismatches, assigning a severity level to each alert. Every flagged issue is supported with clear clinical reasoning, relevant medical references, and identification of the affected medications to enable informed decision making.

Doctor Notification & Escalation: Pharmacists can escalate flagged issues directly to the prescribing doctor, with the option to include a note explaining the potential clinical risk for further review and clarification.

Change Proposal & Clinical Approval: Pharmacists can propose AI-suggested or manually selected alternative medications to the prescribing doctor for review. The doctor can then approve the modification or decline it with documented clinical reasoning.

Final Review & Confirmation: Once all alerts are addressed, pharmacists are presented with a consolidated summary highlighting updated medications in grey and unchanged ones in green, along with complete dosage and prescription details for final verification.

Developed a deeper understanding of real world clinical complexity and the importance of grounding design decisions in contextual user research rather than assumptions.
Learned to design AI systems that are assistive and not replacive by clearly defining decision boundaries, user authority, and ethical constraints.
Strengthened my ability to scope high stakes solutions realistically while balancing innovation with regulatory, safety, and workflow considerations.
Embraced an agile and iterative design approach by revisiting research, testing assumptions, and refining concepts continuously instead of following a linear process.