Revisited Can What Was Once Science Fiction Help Our Growing Rural Health Care Crisis?
Emergency Medical Assistants no longer Fiction
This article was originally published in July 2025. The article was factual and addressed a shortage of medical professionals that has only become worse.
However, it also contained errors and typos.This updated version adds content and corrects any typographical errors, usnder a slightly different title to preserve the original publication date to maintain the timestamp on authorship.
Introduction: A Fading System
There used to be a doctor in every town.
Like a lighthouse — maybe not visible every day, but steady, trusted, there.
Now, one by one, those lights are going out. The Big Beautiful Bill passed by the Republican Majority strips nearly a trillion dollars from health Care, a lot of that comes from Medicaid, and the biggest hit will be already strapped for funds, rural Clinics.
Hospitals are closing. OB-GYNs are disappearing. Rural clinics are understaffed or gone. And even the physicians who remain are under pressure from laws that criminalize care, particularly for women.
This isn’t a gradual erosion — it’s a targeted dismantling. And it's creating a crisis no future administration will be able to reverse quickly. Even with funding, the time to train and deploy professionals remains the choke point.
Here’s a visual representation of the projected Medicaid cuts under the “One Big Beautiful Bill Act” (OBBBA), showing both the spending reductions and the estimated number of people who could lose coverage in specific categories. The dual-axis chart emphasizes how significant financial changes could correlate with sharp drops in coverage.
Science Fiction as Blueprint
On Star Trek, the Emergency Medical Hologram (EMH) was designed to step in when the doctor was unavailable. Not to replace human expertise, but to stabilize, advise, and buy time.
That’s the exact situation we’re entering — not in deep space, but in rural Alabama, Texas, and Nebraska.
What was once science fiction is rapidly becoming a necessity.
Tesseract CDSS: A System Built for Crisis
When I first conceived of Tesseract, I intended it as a methodology, not a design for a hypothetical application. Let's take a look at what a Tesseract system would look like when adopted by the medical profession. This will be a high-level analysis and a paper or more detailed article may be coming.
Lets call this hypothetical application Tesseract CDSS (Clinical Decision Support System). This would be a new class of tool not a chatbot, not a database but a structured assistant explicitly designed for:
Physician Assistants
Clinics operating under strain
It works with EHR systems like Epic. Clinician notes — symptoms, diagnoses, personal observations — become structured prompts. Tesseract CDSS reads these (locally and encrypted), and returns:
Differential diagnoses
Suggested exams and tests
Risk indicators
Access to scoped medical research
A two-panel interface:
Top: clinical suggestions and reasoning
Bottom: actionable follow-ups (clickable for more detail)
Privacy by Design
Tesseract CDSS is built from the ground up for privacy and compliance
The application could include a physician’s notebook patterned after existing systems, such as Epic. Such notebooks have tabs to record the clinician’s diagnosis, overall physical presentation, reported systems, test results, and other relevant information.
And a tab for recording personal observations, such as how the patient presented themselves, their cognitive state, dress, and appearance, as well as their speech (do they articulate sentences clearly without strain).
The Clinical information could even be imported from existing EHR systems.
A deterministic system or local model can extract the relevant parts of the clinical appraisal and send it to the Large Language Model. All PI is removed, just enough info provide the system with an overall picture of the patent’s health state.
Key Features expected of the notebook
All LLM prompts are anonymized
No raw PHI leaves the device
Scoped RAG only accesses:
PDR-equivalent datasets
Peer-reviewed journals
Public health orgs (CDC, WHO, etc.)
Sub-Lex and GROSS
Because sub-lex is lightweight compared to traditional encryption and can even implement role-specific segments, it would be ideal for securing the patient data well within US HIPAA requirements.
more information on Sub-lex and its use in Medical records can be found in a previous article Reimagining medical Records as well as outer posts in this Substack.
Sub-Lex then applies role-based encoding and segmentation.
A PA might only see the top-level summary and suggested differentials.
A consulting specialist might see structured labs and relevant journal citations.
The local model gets de-identified data only — just enough to process context safely.
The summarized anonymized patient record could then be presented to the system perhaps even using the GROSS language I introduced in an earlier conceptual post.
#GROSS:v1.1
context.encounter_id: "rural-clinic-442A"
op:diagnose
subject:patient.anonymous
symptoms: [fatigue, blurred_vision, shortness_of_breath]
labs: { A1C: 8.9, BP: "145/92", LDL: 180 }
risk_flags: [type2_diabetes, cardiovascular_risk]
next_steps: [order:glucose_tolerance_test, refer:endocrinology]
llm_instructions: summarize.differentials | rank:risk_levelBring in the Medical Community
The chat interface should be designed for collaboration. At any time, the clinician should be able to add consulting professionals to the conversation; they can interact with both the PA or clinician using the system or with the LLM itself.
This dual collaboration helps other professionals to consult and critique any recommended treatment programs. Collaboration adds to the chain of custody around the diagnosis.
The model is not the doctor — it’s one voice in the room.
Prerequisite: Fixing RAG:
Retrieval-Augmented Generation. (RAG) is powerful, but in medicine, power without constraint is danger.
What RAG needs is not more freedom, but structured constraint.
And that structure must come from medical institutions, not tech companies.
RAG must begin with a verified, static medical reference layer — a digital analog to the Physician’s Desk Reference (PDR).
This layer is curated, locked, and versioned — the clinical equivalent of “base truth.”
No hallucinations. No Reddit posts. No mix-ins from open web search.
This is the primary lookup layer for dosage, contraindications, triage flags.
Allow Contextual Expansion
On top of this, RAG can extend outward — but only in controlled zones:
Recent peer-reviewed journals
Clinical trial databases
Verified health organization updates (CDC, WHO, ACOG, etc.)
These sources are labeled and timestamped, so the user knows:
“This isn’t doctrine — it’s emerging knowledge.”
Label the Inference Layer
Every RAG result must expose how it reasoned:
What was retrieved
How it was weighted
Which logic pathways were followed
Where uncertainty lies
Medicine tolerates ambiguity — but not secrecy.
Reclaiming the Term "Decision Support System"
CDSS used to mean brittle expert systems from the early 2000s — rigid, checkbox-heavy, and unusable.
But the original expert systems laid the groundwork for what we’re now trying to build:
Historical Use of Expert Systems:
Oil & Gas: managing drilling operations, reservoir analysis, equipment fault detection
Aerospace & Defense: aircraft diagnostics, mission planning, missile guidance
Finance: credit scoring, risk analysis, fraud detection
Manufacturing: process optimization, predictive maintenance, quality assurance
Healthcare (early examples): MYCIN, INTERNIST-I, used for infection diagnosis and internal medicine rule evaluation
These systems were:
Deterministic: given the same input, they always produced the same output
Built for reliability, transparency, and critical decisions
Trusted in high-risk domains where consistency mattered more than creativity
What we have now are:
Flexible, contextual LLMs
Federated learning models
Structured interfaces
Tesseract CDSS would be a marriage of expert system methodology and modern AI. It doesn’t make decisions — it supports them, transparently and repeatably.
From Frames to Vectors: A New Era of Search
Today, we’ve entered a new paradigm: LLM-powered RAG (Retrieval-Augmented Generation) workflows. In this setup:
A private vector store (such as Pinecone, Weaviate, Chroma, or Vertex AI Matching Engine) replaces static databases.
User questions are transformed into vector embeddings, and similar chunks of pre-indexed knowledge are retrieved in real-time.
These relevant chunks are passed to an LLM (like GPT-4 or Claude) to generate context-aware answers—no need to fine-tune the base model.
The loop becomes tighter and more natural: users interact with the LLM, which in turn calls structured search APIs, retrieves semantically relevant results, and generates answers grounded in private or curated data.
This RAG flow supports custom enterprise apps, knowledge retrieval tools, and agent frameworks where hallucination must be minimized and private context matters.
How the LLM Interacts with the Expert System
In traditional systems, only trained professionals could interact with the expert system. Tesseract CDSS changes that by allowing the LLM to act as the intermediary:
The expert system generates a deterministic result set
The LLM analyzes the results, ranks them, and explains them in plain language
The clinician makes the final judgment with more clarity, less guesswork
The LLM doesn’t invent decisions — it clarifies structured ones. It prioritizes, flags ambiguity, and recommends follow-up steps — always with reasoning.
The LLM doesn’t replace the expert — it expands who can access the system.
Traditional Model:
Expert System ←→ Human Expert
Requires domain knowledge to use properly
Rigid inputs, deterministic outputs
Trust and interpretability limited to trained professionals
Tesseract CDSS Model:
Expert System ←→ LLM ←→ Clinician (PA, nurse, GP, etc.)
The LLM translates natural language, fills context, and navigates ambiguity
The Expert System delivers structured, auditable results
The clinician interacts through the LLM-enhanced interface, with suggestions, summaries, and guided explanations
Tesseract as Field Infrastructure
Tesseract CDSS isn’t just for rural clinics. It could be deployed anywhere expertise is thin or delayed:
Ships at sea
Aircraft carriers (EMAS nodes)
Disaster response teams
ISS (or future off-planet installations)
Mobile field hospitals
With proper offline fallback models, Tesseract CDSS can run even in low-connectivity environments offering diagnostic scaffolding where full care isn't available.
The Real Use Case: America as the Emergency Zone
This isn’t just for "developing countries."
We may need to deploy Tesseract CDSS domestically as part of a national medical rescue program.
Red states are losing entire care infrastructures. Laws are forcing clinicians out. Even if future leadership funds reconstruction, the clinicians won’t exist in time.
This is triage at the national infrastructure level.
Global South ≠ Rural America
We must not conflate underserved areas of the U.S. with the Global South.
Global South challenges:
Lack of baseline infrastructure
NGO-driven care
Power/internet scarcity
Tesseract adaptations would include:
Local-only inference
Satellite sync (Starlink, OneWeb)
Solar-charged devices
Multilingual adaptation
It’s possible — but a separate strategy from domestic deployment.
Community + Human Layer
Tesseract CDSS isn’t just an assistant — it’s a hub of shared intelligence.
Doctors, PAs, and nurses can:
Share anonymized cases
See peer commentary in real-time
Flag unusual conditions to the community
Use the AI and the human network together
The LLM doesn't isolate — it joins the room.
Conclusion and call to action
Call to Medical Associations
This is where medical boards and professional associations must step up.
You don’t need to build the model — but you must define the rails it runs on.
A model trained on real science — but fed garbage queries or random web searches — is still malpractice waiting to happen.
Final Wrap: A Methodology for a Medical Emergency
Tesseract is not a product. It is not an app or a brand. It is a methodology — a disciplined, privacy-conscious, role-aware system that binds expert systems, large language models, and human collaboration into something greater than the sum of its parts.
It was born from constraint: the reality that we cannot instantly replace the doctors we’ve lost, the clinics we’ve shuttered, or the trust that has been eroded in rural care.
And now, constraint has become crisis.
The most dangerous myth in healthtech today is that we can “plug in AI” like a new billing system — deploy a chatbot or slap a model onto a search bar and call it transformation. But medicine is not a vertical. It is not a feature. It is a covenant between knowledge, practice, and life itself.
Tesseract doesn’t seek to replace the doctor. It exists to buy time, extend capability, and hold the line until human systems catch up.
It operates within limits:
It refuses to guess without citations.
It cannot answer what cannot be known.
It will tell you where the ambiguity lies — and invite collaboration to resolve it.
Most importantly, it’s designed to function as part of a chain of custody — between human professionals, their peers, and the structured knowledge of clinical science. That chain is what protects patients. And that chain is what Tesseract reinforces, not undermines.
This is not optional innovation.
This may be the only viable response to a collapse that has already begun.
Without it, the lights go out — not just in rural clinics, but across the infrastructure of American medicine.
Tesseract CDSS can work. But only if we build for medicine first — and AI second.
Research Collaboration
If you’re building in the intersection of healthcare, privacy, and AI infrastructure — or working on expert systems for real-world use — I’d love to connect. I’m currently refining this framework for clinical safety and considering early research or investor partnerships.