Long-form document | Project background and full argument.

This page preserves the longer original proposal while the rest of the site provides shorter, role-specific paths for readers who want a faster overview.

Contact

THE JOY PROJECT

SECE-MedAssist

Synthetic Emotional Core Engine for Medical & Assistive Intelligence

A Proposal for Expert Evaluation and Research Support

Paul Totoritis | 2026 | Zenodo: Advancements in Artificial Intelligence for Medicine

Abstract

SECE-MedAssist (Synthetic Emotional Core Engine for Medical and Assistive Intelligence) is an experimental hybrid analog-digital cognitive architecture designed to address a critical gap in current medical AI: the absence of genuine emotional context, persistent memory, and privacy-first design.

The system consists of two tightly integrated layers. The hardware layer uses analog op-amp integrator circuits to physically model internal emotional states — not as software variables, but as real voltages that accumulate, decay, and interact over time. The software layer applies an Emotional Modeling Engine (EME) that uses sigmoid-curve slope calculation to modulate AI response tone based on real-time patient distress, storing all interactions in a fully local Portable Knowledge Base.

This document presents the architecture, novelty, current development status, and the case for research support to bring SECE-MedAssist to clinical validation.

1. The Problem This Solves

Current medical AI systems, however capable, share three fundamental deficiencies:

No emotional context. Medical AI responds to clinical inputs but is blind to the patient’s emotional state. A patient expressing fear, grief, or confusion receives the same output as a calm, compliant patient. This produces medically accurate but humanly inadequate responses.
No meaningful memory. Interactions start fresh each session. The system cannot recall that a patient was anxious last week, that they responded better to encouragement than instruction, or that their fear is escalating. Continuity of care is impossible without continuity of memory.
No privacy sovereignty. Most AI systems require data to transit to cloud servers for processing, creating HIPAA compliance complexity, patient distrust, and data breach exposure. Patients and clinicians have no true control over where patient data lives.

These are not minor interface problems. They reflect a fundamental architectural mismatch: medical AI is built on systems designed for information retrieval, not for the kind of sustained, emotionally-aware relationship that defines good care.

2. The SECE-MedAssist Solution

2.1 Core Concept

SECE-MedAssist is described in its own documentation as a “minimal synthetic nervous system.” This is not metaphor. The system is designed to develop internal preferences and behavioral tendencies through experience — just as a nervous system does.

It is not a chatbot. It is not a rule-based decision tree. It is a cognitive architecture in which emotional state is a physical quantity, not a computed label.

2.2 The Hardware Layer

The analog circuit — documented in the SECE_MedAssist.net schematic and implemented in KiCad as the Pentad Cortex PCB — operates as follows:

Three environmental sensors (TMP36 temperature, microphone, light-dependent resistor) feed into LM358 operational amplifier non-inverting stages.
Each amplified signal drives a CD4046B voltage-controlled oscillator (VCO), converting analog amplitude to frequency — a form of analog encoding that naturally captures signal dynamics.
Three VCO outputs combine in a summing amplifier. The mixed signal then drives four integrator circuits: State A (fast, 1µF), State B (medium, 10µF), State D/Attention (10µF), and a slow Memory integrator (100µF).
States are cross-coupled: State A influences State B, and the Memory state feeds back into both, creating emergent dynamics that cannot be reduced to simple lookup tables.
All state voltages are level-shifted from the 5V domain to the ESP32’s 3.3V ADC range using resistor dividers. The ESP32 microcontroller reads all states and outputs control signals via I2C to the MCP4725 DAC.
A reward potentiometer provides physical reinforcement input. A learning-enable switch gates whether the current reward signal modifies the system’s behavior.

The significance of this hardware approach is that emotional state is not computed — it is measured. The integrators accumulate charge over time, representing the accumulation of experience. This is qualitatively different from maintaining a numerical variable in software memory.

2.3 The Software Layer

The Emotional Modeling Engine (EME) implements the following:

Weight = Intensity × Clarity × Context_RelevanceSignal Weight Formula:
A sigmoid function calculates the ‘momentum’ of patient emotion. Rather than classifying distress as binary (high/low), the system models how distress builds and recedes over time.Analog Slope Modeling:
The system maintains a baseline bias of 0.6 toward supportive response. Above this threshold, the system defaults to empathetic tone; below, to direct clinical fact delivery.Positivity Bias:
Each emotional signal carries a persistence_score that modulates its decay rate. Positive memories can be reinforced by self-check (10% boost per cycle) or locked permanently by supervisor approval (MAX_PERSISTENCE = 1000). Negative memories can be manually cleared — a design choice that models therapeutic forgetting.Memory Persistence Scores:
All patient interactions are stored in a local JSON file. There is no cloud dependency. Patient history — including emotional resonance scores — remains on the clinical device.Portable Knowledge Base:

2.4 Integrated Operation

When a patient says “I am scared I won’t walk again” with a distress level of 0.9:

The clinical reasoning component retrieves the medically accurate response: 6 weeks recovery with daily physical therapy.
The EmotionalWeightEngine calculates a slope from the distress value. Slope > 0.5 triggers supportive mode.
Output: “I understand this feels like a long road, but [clinical fact]. We will monitor your progress every step of the way.”
The interaction is logged locally with its emotional resonance score. No data leaves the device.

3. What Makes This Novel

Dimension	Standard Medical AI	SECE-MedAssist
Emotional State	None — text only	Analog voltage states via RC integrators
Memory	Session-only or cloud	Persistent local PKB with configurable decay
Response Tone	Fixed or template-based	Sigmoid-modulated: supportive ↔ direct
Privacy	Cloud-dependent processing	Fully on-device, patient-owned data
Learning	Periodic retraining only	Real-time reward signal via potentiometer
Physical Substrate	Digital software only	Hybrid analog-digital — embodies the model
Memory Management	Accumulates indefinitely	Decay + reinforcement + therapeutic deletion

To the author’s knowledge, SECE-MedAssist is the only system that implements emotional state as physical analog computation in a medical AI context. Related work in neuromorphic computing (e.g., Intel’s Loihi chip) pursues analog computation at the silicon level, but not for emotional state modeling in clinical applications. The Joy Project represents an accessible, low-cost entry point into this design space.

4. Current Development Status

Completed

Full analog circuit designed as SKiDL (Python) netlist — multiple validated iterations
KiCad PCB layout complete — Pentad Cortex board ready for fabrication
EmotionalWeightEngine implemented and tested in Python
SECEMedAssist architectural prototype built and executed in Kaggle/Jupyter
Portable Knowledge Base implemented with local JSON storage
Memory decay, persistence scoring, and reinforcement simulation complete
Proposal submitted to Zenodo community: Advancements in AI for Medicine
Complete text schematic documented for reproducibility
Emotional simulation archive with interactive I×C×R resonance engine

Immediate Next Steps

PCB fabrication and hardware assembly of Pentad Cortex
ESP32 firmware: ADC state reading + I2C DAC control loop
Integration of analog hardware states with Python software layer
Clinical scenario validation with simulated and real patient data
Peer-reviewed publication in AI in Medicine or Nature Digital Medicine
IRB application for pilot study in senior care or rehabilitation setting
Open-source release of full codebase, schematics, and documentation

5. Request for Support

The Joy Project is at a pivotal stage. The theory is validated. The circuit is designed. The software prototype runs. What remains is translation from design to working system, and from prototype to clinical evidence.

Support is sought in the following areas:

Hardware fabrication: PCB manufacturing and component procurement for the Pentad Cortex prototype ($500–$2,000 estimated for initial batch).
Clinical collaboration: Access to a clinical environment for scenario testing, IRB guidance, and patient interaction protocols.
Research mentorship: Engagement with qualified researchers in AI, neuroscience, or medical informatics who can provide peer review and co-authorship on formal publication.
Institutional affiliation: Formal association with a research institution to facilitate grant applications, IRB approval, and academic publication.
Technical review: Expert evaluation of the analog circuit design, firmware architecture, and software emotional modeling for correctness and safety.

The Joy Project is not asking for belief that this will work. It is asking for the opportunity to find out — with the rigor that question deserves.

6. Conclusion

Medicine has always required two things that exist in tension: clinical precision and human connection. Every good physician knows that the right fact delivered without empathy can do harm; the wrong tone at the wrong moment breaks trust that took years to build.

SECE-MedAssist is an attempt to give AI the architecture it needs to hold both at once — not by programming empathy in as a rule, but by building a system that develops it through experience, the way any mind does.

The hardware is designed. The software runs. The vision is clear. What is needed now is the partnership of qualified professionals who can help determine whether this vision is sound — and if so, help bring it into the world.

Paul Totoritis | SECE-MedAssist / The Joy Project | 2026

Zenodo Community: Advancements in Artificial Intelligence for Medicine

Research Proposal