The OrthoSystem Vision

OrthoSystem didn't emerge from a startup accelerator or a corporate AI lab. It emerged from the unusual intersection of two careers that shouldn't have overlapped—and the realization that understanding both domains deeply enough reveals solutions invisible to specialists in either.

OrthoSystem integrates four complementary subsystems to address fundamentally different modes of knowledge access. The platform recognizes that experts and learners approach information differently—and provides tools optimized for each cognitive style while leveraging shared infrastructure.

When you ask ChatGPT a question, it generates an answer from patterns learned during training—patterns frozen at whatever date the model was last updated. It has no access to your documents, your textbooks, or the latest research. Retrieval-Augmented Generation (RAG) changes this. Before the language model generates a response, a retrieval system searches your documents and injects the relevant passages into the prompt. The model then generates an answer grounded in actual sources rather than statistical memory.

The concept sounds simple. The implementation is not.

Why Existing Solutions Failed Us

Our initial exploration of available RAG platforms revealed fundamental limitations. Many commercial solutions impose arbitrary file size limits—cutting off textbook chapters mid-sentence. Others use generic embedding models trained on web text, poorly suited to medical terminology where "reduction" means something entirely different than in everyday English. The retrieval quality varied wildly depending on how questions were phrased.

The deeper problem is architectural. RAG isn't a single tool—it's an orchestration of multiple AI subsystems, each making decisions that compound downstream.

Each component is itself a "mini AI"—a specialized neural network making decisions that affect final answer quality. The embedding model casts a wide net on what's similar. The chunker decides what constitutes a coherent unit. The reranker decides what's actually relevant. Commercial platforms make these decisions for you, generically. Building OrthoRAG meant taking control and fine tuning each stage.

Two Parallel Systems: V33 and V41

We maintain two parallel RAG implementations, each serving different reliability and capability requirements.

V41's additional metadata enables OrthoGraph's multiple semantic views—organizing the corpus by subspecialty, anatomy, or evidence level requires that information to exist in the database.

When I search the literature, I have a precise question—"What's the latest evidence on graft selection for ACL reconstruction in adolescents?" I'm filling a specific gap in my knowledge.

Learners don't have that framework yet. It’s not as if they're standing at the edge of unfamiliar territory with no map. That giant textbook? Learners don’t need help figuring out what parts are important — everything is important. Breaking down a massive universe of information into learnable chunks when residents are working 70-80 hours a week in the hospital while trying to preserve healthy work-life balance and personal commitments is the challenge.

This insight drives the architecture. We need multiple different systems because we all learn in different cognitive modes.

The Origins of Word Vectors

The mathematics that make this possible emerged not from Silicon Valley but from university research. In the 1950s, Charles Osgood and his team at the University of Illinois asked hundreds of undergraduate students to rate concepts—words like lady, father, fire, boulder, tornado, and sword—on fifty different scales: good-bad, hard-soft, hot-cold, sweet-sour. Across multiple studies and populations, concepts consistently clustered into three dimensions: good-bad, strong-weak, and active-passive. Each word could be described by three numbers—just like x, y, z coordinates. This research utilized vacuum tube supercomputers from IBM.

The next major leap did come from Silicon Valley, when Mikolov hypothesized that the meaning of a word is defined by the words it tends to occur with in 2013. Using a Google News dataset with six billion words, his team placed one million common words in 300 dimensions—derived purely from patterns of co-occurrence. The stunning result: if you took the coordinate for king, subtracted man, and added woman, the nearest word was queen. Paris minus France plus Italy? Rome. Meaning had become geometry.

In 2017, the transformer architecture enabled this mathematical structure to scale to entire sentences and documents. OrthoGraph leverages this same geometry—the embeddings that power retrieval become coordinates in visual space. Semantic similarity becomes spatial proximity. A search doesn't return a list; it illuminates a region.

Hardware Requirements

The DGX Spark's 128GB unified memory enables all services to coexist on a single node for development. Production deployments with larger models benefit from dual-node configurations where inference runs on dedicated hardware. More sophisticated models have more sophisticated reasoning capabilities.

Every architectural decision reflects a constraint or an insight. Microservices add complexity but enable scaling as the technology grows. Qdrant gets selected among a multitude of vector databases because it targets our latency priorities and enable storage of our 768-dimensional projections alongside 4096-dimensional embeddings without a separate database.

Learners face a challenge that experts forget: when you don't know a domain, you don't know what questions to ask. OrthoGraph addresses this by transforming the corpus into navigable space. The same vector embeddings that power retrieval become coordinates in a visual field where semantic similarity becomes spatial proximity.

A search doesn't just return a list—it illuminates a region. Clusters reveal topic structure. Sparse areas indicate gaps. The learner develops intuition about the knowledge landscape before diving into specific content.

Technical Implementation

Real-time visualization of 50,000+ chunks at 60fps requires careful engineering. CPU-based UMAP on 50K vectors at 4096 dimensions takes 15-30 minutes. GPU UMAP via cuML completes in under 2 minutes.

Multiple Semantic Views

The same data projects differently depending on what structure you want to see. Switching views triggers smooth interpolation through intermediate positions—the corpus "morphs" from one organization to another.

Service Architecture

RAG Compatibility Tiers

OrthoGraph adapts to different OrthoRAG deployment levels. The tier is auto-detected on startup by probing Qdrant payload fields.

Ports are organized by subsystem. Internal services communicate via Docker network DNS and are not exposed to the host.

OrthoRAG (V41)

OrthoGraph

OrthoTutor

Each subsystem follows single-script deployment. Scripts include dependencies, health checks, and graceful error handling. The RAG_PREFIX environment variable enables switching between V33 and V41 deployments.

orthograph-api:
  image: nvcr.io/nvidia/pytorch:25.11-py3
  networks:
    - ortho-network
    - ${RAG_PREFIX:-v34}_rag-network
  environment:
    - RAG_PREFIX=${RAG_PREFIX:-v34}
    - QDRANT_HOST=${RAG_PREFIX:-v34}_qdrant
    - QDRANT_COLLECTION=${QDRANT_COLLECTION:-documents_multiscale}

orthograph-projection:
  image: nvcr.io/nvidia/pytorch:25.11-py3
  deploy:
    resources:
      reservations:
        devices:
          - driver: nvidia
            count: 1
            capabilities: [gpu]

networks:
  ortho-network:
    driver: bridge
  ${RAG_PREFIX:-v34}_rag-network:
    external: true

Memory Configuration

OrthoRAG (V41) and OrthoBench are operational. The immediate focus is OrthoGraph—the visual corpus navigation that will demonstrate the platform's educational potential. OrthoTutor, the agentic system that ties everything together, represents the largest development effort and the primary funding opportunity.

About the Author

Alan B.C. Dang, MD

Health Sciences Clinical Professor of Orthopaedic Surgery, UCSF
Staff Surgeon, San Francisco VA Health Center

Three decades in computing. A decade in academic spine surgery. Three patents. One vision: AI that teaches the way surgeons actually learn.

UCSF Faculty Profile · LinkedIn