Projects
The Architect
Production (PyPI)Public • Open-Source
- An autonomous development lifecycle framework for agentic AI coding tools that plans, executes, retries, reviews, and retains knowledge without supervision.
- Maintains persistent project memory in ARCHITECT.md, capturing design decisions, constraints, and lessons across sessions.
- Employs multi-signal completion detection and circuit breaker mechanisms to prevent agent hang-ups and hallucinated outputs.
- Supports headless and CI modes alongside a live tmux dashboard for versatile deployment workflows.
- Available on PyPI, installable via pip, with compatibility for Claude Code and OpenCode AI models.
MHIL
Concept & Early Prototyping (TRL-5)Semi-Stealth • Non‑Profit
- Machine Human Intelligence Lab is a non-profit research institute focussed on creating advanced benchmarks and evaluation protocols that measure genuine machine intelligence beyond narrow LLM benchmarks.
- Constructs human-aligned test suites evaluating agentic capabilities including memory, planning, robustness, and autonomous problem-solving under uncertainty.
- Develops open evaluation matrices applicable to pre-superintelligent systems for standardized, vendor-agnostic, architecture-independent comparisons.
- Establishes future-focused standards with reproducible scorecards, capability tiers, and safety-critical thresholds essential for real-world superintelligence deployment.
MazeByte
SOTA Proof of Concept (PoC)Semi-Stealth
- A cutting-edge autonomous AI agent prototype for fully self-maintained ETL and data engineering automation.
- Integrates infant-like feedback loops based on reward and pain signals alongside resource-seeking behaviours for continual adaptive learning.
- Generates modular pseudo-code components managing memory, search, optimisation, and orchestration to automate the entire ETL lifecycle.
- Utilises the proprietary MBI internal model to replace manual data pipeline maintenance with scalable agentic data systems.
SightBit
SaaS (TRL-9)Public
- A production-grade computer vision SaaS platform delivering real-time drowning risk prediction, human and vessel detection, and flood-risk alerts using standard CCTV streams without specialised sensors, edge hardware, or manual site calibration.
- Employs a panoptic segmentation deep learning model augmented with a proprietary pipeline that significantly improves recall of small objects by 18 percentage points.
- Designed for operational reliability and scalable deployment in safety-critical aquatic and flood-risk monitoring environments.
AdaptiveBridge
Production-ready (TRL-8)Public • Open-Source
- Open-source Python library engineered for production ML robustness by predicting and filling missing features at inference time, preventing common production failures from incomplete data.
- Features configurable thresholds for feature importance, correlation handling, and supports custom accuracy metrics with automated distribution adjustments.
- Implements a drop-in estimator interface with fit and transform methods for seamless integration into existing ML pipelines.
- Explicitly manages complex feature dependencies, including mandatory, deviation, and leveled relationships to optimise model performance under missing data scenarios.
- Delivered with comprehensive documentation, unit tests, continuous integration, and licensed permissively on PyPI.
Jhive
Production (TRL-7)Public • AIforGood
- A Django-based, AI-for-Good web application leveraging the OpenAI API to perform real-time detection and classification of antisemitic content.
- Provides a rich reporting interface for content analysis and review.
- Offers secure, RESTful API endpoints featuring rate limiting for controlled integration.
- Implements rigorous input validation and sanitisation protocols to guard against injection attacks and ensure operational security.