I Built a Healthcare Platform Where AI Is Never Allowed to Have the Final Word.

Diagnosia: Connected Care, Reviewable by Design

Project Overview

Healthcare information is dangerously fragmented across brief consultations, handwritten prescriptions, laboratory documents, hospital discharge papers, and specialist referrals. To bridge these critical gaps, I developed Diagnosia.

Rather than a collection of disconnected AI demonstrations, Diagnosia is a cohesive healthcare workflow platform designed to help clinicians, specialists, patients, and care teams manage information seamlessly across the entire patient journey.

The system operates on a coordinated workflow where:

1. Information enters through existing channels.
2. AI prepares reviewable drafts for medical professionals.
3. Authorized modules reuse relevant context.
4. Workflow events dynamically update the patient journey in real time.

Core Capabilities & Modules

The platform is anchored by three foundational tools:

• DiagnoFast: A voice-first patient intake system.
• Skin Cancer Detection: Image-based screening support.
• Prescription Analyzer: Medication interpretation.

To extend this foundation, four specialized modules provide additional support:

• ClinScribe: Clinical encounter summarization.
• LabLens: Laboratory report digitization.
• DischargeCare: Transition coordination for hospital discharges.
• ReferralLoop: Specialist referral tracking.

Demonstration & Patient Journey

View the demonstration files on Google Drive

The platform's visual identity emphasizes operational clarity, demonstrating an end-to-end synthetic patient journey on a live timeline. The workflow captures a voice interview to queue a report, digitizes scanned tests, and drafts clinical notes based entirely on consultation transcripts. Every state change is projected live onto the timeline without requiring manual screen reloads.

Technical Architecture & Partnerships

Diagnosia integrates several partner technologies, each with a strictly defined and isolated responsibility to prevent conflicting sources of truth:

• Google Gemini API (Multimodal Engine): Reads scanned lab reports and prescription photos. All outputs are validated against a strict schema before reaching human review.
• YugabyteDB (Technical Audit Trail): Maintains a record of all AI runs, attempts, and handoffs, keeping the core clinical database clean.
• Meko via MCP server (Shared Agent Context): Manages approved knowledge and memory.
• MongoDB Atlas (Workflow Event Store): Uses Change Streams to instantly project workflow events onto open care screens.

Architecture Principles & Safety Boundaries

The engineering principles prioritize patient safety and data integrity above all else:

• Single Responsibility: No sponsor database is permitted to override the local identity authority, which remains the absolute single source of truth.
• Human-in-the-Loop: All AI outputs are treated strictly as drafts until approved by an authorized human reviewer. Nothing reaches a reviewer unchecked.
• Data Separation: Technical audits are completely separated from clinical data, ensuring records never pollute the core database.
• Strict Authorization: Modules only access context they are explicitly authorized to view.
• Testing Boundaries: As a pilot project, the platform operates entirely on synthetic data.

Project Philosophy

Building a modular, feature-frozen product with strict safety boundaries was an incredible test of technical discipline. Wrangling the architecture to ensure the local identity authority remained secure reinforced a core lesson: building for healthcare is not just about rapid innovation; it is about meticulously engineering trust.