<!-- personal homepage -->
dominic
title: PhD Researcher
focus: Persistent Cognition Systems
status: Building
<a href="#research">publications</a>
<a href="#about">curriculum vitae</a>
<a href="#brain">architecture</a>
<a href="#contact">contact</a>
Last updated: March 2026
Persistent
Cognition
Calibration
22.6 / 25
Drift
0.003
uptime:
homeostasis: nominal
Memory, doctrine, enforcement, continuity. The gap between stateless chat and a mind that compounds.
<!-- about -->
Curriculum Vitae
Education
Ph.D. Cognitive Science
2020 – Present
Research Interests
- - Persistent cognitive architectures
- - Memory crystallisation and doctrine enforcement
- - Calibration drift in personalised AI systems
- - Executive control and intent routing
- - Human-AI continuity across time
Experience
Independent Researcher
Cognitive Systems, 2020 – Present
Strategy & Technology
Various, 2018 – Present
Skills
Python, PyTorch, LangChain, Neo4j, FastAPI, Cognitive Modelling, Systems Architecture, Technical Writing
<a href="cv.pdf">Download full CV (PDF)</a>
Why build a cognitive substrate?
The question isn't whether AI can remember you. It's whether it can build a coherent model of you that sharpens over time.
Most AI systems are stateless interfaces layered on raw intelligence. The missing layer is governance: memory with doctrine, calibration with enforcement, continuity with intent.
A persistent cognitive operator doesn't just store context. It predicts, acts, compares, and updates. The architecture work provides precision. The building provides truth.
"The goal is not to store more chat history. It's to build a cognitive substrate that gets measurably sharper at reading you, challenging you, and acting for you — over time."
<!-- publications -->
Publications
Towards Persistent Cognitive Operators: Memory, Doctrine, and Enforcement in Long-Horizon AI
arXiv preprint, 2025
[preprint]
Calibration Drift in Personalised AI Systems: Detection, Attribution, and Correction
Workshop on Human-AI Alignment, 2024
[workshop]
Binder-Based Executive Control for Multi-Module Cognitive Architectures
Conference on Cognitive Systems, 2025
[under review]
<a href="scholar">Google Scholar</a>
<a href="orcid">ORCID: 0000-0000-0000-0000</a>
What I'm building
NanoBot
Persistent cognitive operator with memory crystallisation, homeostasis, doctrine enforcement, and calibration-driven response shaping.
Binder
Executive function layer routing competing module inputs into a single coherent response. Inspectable attribution traces from day one.
Calibration Ledger
Prediction accuracy tracking per cognitive domain. Measures how well the system models the user over time.
Routing quality has repeatedly proven more important than adding more raw intelligence.
<!-- architecture -->
System Architecture
Overview
NanoBot is a persistent cognitive architecture — not a chatbot. It maintains a durable model of the user across sessions, calibrates to individual cognitive patterns, and self-corrects through closed feedback loops. Core loop: predict, act, compare, update.
Core Components
- Memory Store: Crystallised doctrine + mutable live state
- Binder: Executive function layer for competing module inputs
- Homeostasis: Live internal state — uncertainty, urgency, cognitive load
- Metacognition: Critic layer scoring and rewriting weak outputs
- Calibration Ledger: Prediction accuracy by domain, hit/miss tracking
- Drift Engine: Commitment monitoring and identity drift detection
Technical Stack
Python / PyTorch / LangChain / Neo4j / FastAPI / Custom calibration protocols
<a href="docs">Read the technical documentation</a>
Hover over nodes to explore the architecture
<!-- contact -->
Get in Touch
email: dominic@dominic.ac
twitter: @dominic
github: github.com/dominic
<!-- I read all emails. Response time: ~48h -->