Homzdoctor

Problem Statement

Healthcare information is fragmented across uploads, notes, and patient conversations. Patients need understandable guidance, while clinicians need a reviewable record instead of opaque or overconfident AI output. A healthcare assistant must also protect ownership boundaries and avoid turning suggestions into unsupervised treatment.

Proposed Solution

Homzdoctor is a local-first healthcare copilot prototype for patients and providers. It organizes medical uploads, produces clearly labeled research and decision-support results, and routes them through clinician review. The system is designed to keep a human accountable at every point where a medical decision or prescription could be affected.

Innovation and Technology

The platform combines a Python API, a TypeScript web interface, SQLite persistence, scoped retrieval, and optional OpenAI-compatible local models. The default workflow is deterministic and offline, so the core demo needs no external model, database, Redis server, vector database, or API key. AI results are persisted as pending review rather than treated as diagnoses.

Target Users and Potential Impact

Patients can organize their own records and receive clearer, bounded explanations. Clinicians can review structured results and approve or reject the next step. The potential impact is safer access to understandable health information, less fragmented handoff, and better visibility into uncertainty without claiming autonomous diagnosis or treatment.

Implementation Approach

The repository separates backend, frontend, machine-learning, documentation, and infrastructure concerns. Uploads are private, size-limited, extension-validated, and stored under generated names. Patient retrieval is scoped to curated knowledge and the authenticated patient’s records. Doctor and admin access is not self-service; prescriptions require a doctor-reviewed record and clinician approval; pharmacy ordering also requires patient confirmation. Missing optional models produce explicit fallback states.

The project includes tests for storage, authentication and ownership boundaries, clinician review gating, persistent analysis results, retrieval privacy, local-model compatibility, and safe offline agent behavior. It is a local research/demo system and has not been validated for clinical use, regulatory compliance, or autonomous diagnosis or treatment.

Team Details

Solo project by Mina Gayid.

Built With

• ai
• copilot
• fastapi
• healthcare
• python
• react
• sqlite
• telemedicine
• typescript