MedFolder AI Overview

MedFolder AI is a local-first research prototype for turning medical PDFs into structured, reviewable intelligence.

It is designed for technical evaluation and research workflows, not for diagnosis, treatment, or routine clinical use.

Screenshots

Processing workflow

Structured review view

Evidence viewer

Why it matters

Medical information is often scattered across PDFs from different dates and institutions.
MedFolder AI explores how these records can be transformed into a more structured, reviewable workspace for research and technical evaluation.

Core capabilities

• Medical PDF intake and parsing
  Accepts medical PDF documents and prepares them for structured downstream processing.

• OCR-assisted document recovery
  Supports OCR-based handling for noisy, scanned, or imperfect source material.

• Structured medical extraction
  Extracts review-oriented signals such as diagnoses, medications, temporal references, and supporting evidence spans.

• Hybrid pipeline processing
  Combines deterministic logic, heuristic extraction, normalization, confidence scoring, and optional model-assisted stages instead of relying on a single black-box model.

• Cross-document timeline fusion
  Relates findings across multiple documents and dates to support longitudinal review where available.

• Evidence-linked review surfaces
  Organizes extracted findings into inspectable structures rather than presenting opaque final outputs only.

• Conflict and consistency review
  Highlights ambiguity, mismatch, overlap, and case-level review signals across documents and structured findings.

• Local-first workspace design
  Built around privacy-sensitive local processing rather than routine remote handling of raw medical source files.

• Security-conscious processing model
  Explores stronger safeguards beyond local-first handling alone, including controlled data flow, fail-closed contracts, and protected storage and export paths where applicable.

Architecture overview

MedFolder AI is designed as a local-first document intelligence workspace for technical evaluation and research use.

At a high level, the system consists of several cooperating layers:

1. Document intake layer
   Handles PDF upload, parsing preparation, OCR recovery paths, and source-document preprocessing.

2. Hybrid extraction pipeline
   Converts raw document content into structured intermediate signals using a mix of deterministic parsing, heuristic rules, normalization logic, confidence scoring, and optional ML-assisted stages.

3. Medical evidence graph
   Connects extracted findings to supporting evidence so diagnoses, medications, and related signals remain more reviewable and traceable.

4. Cross-document fusion and timeline layer
   Aligns findings across records and dates to support longitudinal review rather than document-by-document reading only.

5. Cubes-oriented systems layer
   Introduces a modular cube-style execution model for stricter boundaries between processing responsibilities, typed envelopes, and more controlled routing across subsystems.

6. Meta-optimization and orchestration layer
   Explores higher-level orchestration strategies for improving extraction behavior, confidence handling, conflict processing, and workflow routing across the system.

7. Workspace UI layer
   Presents documents, evidence, review state, conflicts, and next-step workflow guidance in an inspectable interface rather than a black-box result screen.

Systems and infrastructure ideas

MedFolder AI is not built around a single end-to-end model.
Its direction is closer to a hybrid medical document intelligence stack with several interacting components and architectural safeguards.

This includes work on ideas such as:

• hybrid extraction pipelines
• medical evidence graphs
• cross-document reasoning
• timeline fusion
• conflict and consistency detection
• meta-optimizer concepts for case-level orchestration
• cube-style execution boundaries
• Arrow-oriented handling for larger tabular data paths
• architect-layer style evaluation and workflow control
• local-first review surfaces for human inspection

Privacy and security direction

MedFolder AI is designed around a local-first and privacy-sensitive workflow.

The project direction also includes stronger technical safeguards beyond the local-only default where applicable, such as:

• restricted data movement
• controlled execution boundaries
• fail-closed contract thinking
• safer storage and export paths
• encryption-oriented handling for sensitive artifacts
• structured separation between processing layers

Where needed, protected local storage and artifact handling may include encryption-oriented mechanisms such as AES-based protection for sensitive data paths or retained outputs.

Design principles

• Local-first by default
• Privacy-sensitive by design
• Human review before trust
• Explainability over black-box presentation
• Structured evidence over raw document chaos
• Hybrid pipeline thinking over single-model dependence
• Security-aware architecture, not just UI-level privacy language
• Research prototype, not clinical automation

Current scope

MedFolder AI is currently a research prototype.
It is intended for demonstration, technical evaluation, and exploratory research workflows.

The public repository provides a project overview only.
The main implementation remains private at this stage.

Important notice

MedFolder AI is not a medical device.
It is not intended for diagnosis, treatment, clinical decision-making, emergency use, or routine patient care.

Outputs may be incomplete, inaccurate, or misleading and always require qualified human review.

Planned public additions

• additional screenshots
• feature walkthrough
• development roadmap