Projects
The Architect
Production (PyPI)Public • Open-Source
- Autonomous development lifecycle framework for agentic AI coding tools that plan, execute, retry, review, and retain knowledge without human supervision.
- Maintains persistent project memory in ARCHITECT.md, capturing decisions, constraints, and learned lessons across sessions.
- Employs multi-signal completion detection and circuit breaker mechanisms to prevent agent stalls and hallucinated outputs.
- Supports headless operation, CI integration, and live tmux dashboards for versatile deployment scenarios.
- Available on PyPI, compatible with Claude Code and OpenCode environments.
MHIL
Concept & Early Prototyping (TRL-5)Semi-Stealth • Non‑Profit
- Machine Human Intelligence Lab (MHIL) is a non-profit institute developing rigorous benchmarks and evaluation protocols to measure comprehensive machine intelligence beyond narrow LLM task performance.
- Constructs human-aligned test batteries for agents, tool-using systems, and composite AI architectures, assessing memory, planning, robustness, and autonomous problem-solving under uncertainty.
- Creates open evaluation matrices enabling standardized, vendor-agnostic comparisons across models and architectures.
- Focused on establishing future-proof, "superintelligence-ready" standards featuring reproducible scorecards, capability tiers, and safety-critical thresholds for real-world deployments.
MazeByte
SOTA Proof of Concept (PoC)Semi-Stealth
- State-of-the-art proof of concept demonstrating a fully autonomous ETL agent driven by self-maintaining AI capable of managing end-to-end data pipelines without human intervention.
- Integrates infant-inspired feedback loops using reward and pain signals alongside resource-seeking and self-preservation behaviours for adaptive, continuous learning.
- Generates dynamic pseudo-code modules handling memory, search, optimisation, and orchestration to automate scalable data engineering tasks.
- Built around a proprietary MBI model, MazeByte aims to replace manual ETL workflows with robust agentic automation.
SightBit
SaaS (TRL-9)Public
- Production-grade computer vision platform delivering real-time detection and prediction for drowning risk, human and vessel presence, and flood hazards.
- Operates solely on standard CCTV video feeds without dependence on specialised sensors, edge devices, or manual per-site configuration.
- Utilises an advanced panoptic segmentation model augmented by a proprietary data augmentation pipeline, resulting in an 18 percentage point increase in small-object recall critical for safety monitoring applications.
AdaptiveBridge
Production-ready (TRL-8)Public • Open-Source
- Open-source Python library designed to maintain machine learning model robustness by predicting and imputing missing features dynamically during inference, addressing a common failure mode in production ML systems.
- Features configurable thresholds for feature importance, correlation, and supports custom accuracy logic with automated distribution adjustments.
- Provides a drop-in estimator interface supporting the standard fit-transform pipeline, enabling seamless integration with existing ML workflows.
- Manages complex feature dependency hierarchies including mandatory, deviation, and leveled features to optimise inference stability.
- Ships with comprehensive documentation, unit tests, continuous integration, and is released under a permissive license on PyPI.
Jhive
Production (TRL-7)Public • AIforGood
- Django-based production web application leveraging the OpenAI API to detect antisemitic content in real time.
- Implements a robust content classification engine paired with an intuitive reporting interface for analysis and review.
- Exposes RESTful API endpoints with built-in rate limiting for scalable integration.
- Enforces secure input validation and rigorous sanitisation protocols to prevent injection vulnerabilities and misuse, tailored specifically for antisemitism detection in sensitive online environments.