Projects
The Architect
Production (PyPI)Public • Open-Source
- Autonomous AI coding lifecycle framework that plans, executes, reviews, retries, and retains knowledge without supervision.
- Implements persistent project memory through ARCHITECT.md, capturing decisions, constraints, and lessons for continuity across sessions.
- Incorporates multi-signal completion detection and circuit breakers to prevent agent deadlocks and hallucinated outputs.
- Supports headless continuous integration workflows alongside interactive tmux dashboard monitoring.
- Available as a PyPI package, installable via pip, compatible with Claude Code and OpenCode environments.
MHIL
Concept & Early Prototyping (TRL-5)Semi-Stealth • Non‑Profit
- The Machine Human Intelligence Lab is a non-profit research entity creating rigorous benchmarks and evaluation protocols to assess genuine machine intelligence beyond narrow natural language tasks.
- It designs human-aligned test suites covering key agent capabilities including memory retention, strategic planning, robustness under uncertainty, and autonomous problem-solving.
- MHIL develops an open evaluation matrix for comparing ‘‘pre-superintelligent’’ AI systems across architectures and vendors on a level playing field.
- Focused on establishing reproducible measurement scorecards, capability tiers, and safety-critical thresholds to guide responsible real-world AI deployment standards.
MazeByte
SOTA Proof of Concept (PoC)Semi-Stealth
- Pioneering fully autonomous AI agents engineered for self-maintained, end-to-end data engineering and ETL automation.
- Leverages infant-inspired feedback mechanisms based on reward and pain signals to drive adaptive learning and resilient resource-seeking behaviours.
- Generates modular pseudo-code components enabling memory management, search operations, optimisation routines, and orchestration of complete data pipelines.
- Employs a proprietary MBI internal model to eliminate manual intervention throughout the data lifecycle, advancing agent-centric automation for complex data systems.
SightBit
SaaS (TRL-9)Public
- Production-grade computer vision platform delivering real-time detection of drowning risks, human and vessel presence, and flood hazard alerts using standard CCTV feeds.
- Operates without reliance on specialised sensors, edge computing nodes, or site-specific calibration.
- Employs a panoptic segmentation deep learning model augmented by an in-house data augmentation pipeline, achieving significant gains in detecting small-sized objects critical for safety monitoring and risk prediction.
AdaptiveBridge
Production-ready (TRL-8)Public • Open-Source
- Open-source Python library engineered to enhance machine learning inference robustness by predicting and imputing missing features dynamically in production.
- Supports configurable feature importance thresholds, correlation-aware fill strategies, and custom accuracy logic to adapt to real-world data variability.
- Offers a drop-in transformer interface integrating seamlessly with existing ML pipelines via fit and transform methods.
- Manages complex feature dependencies—mandatory, deviation-based, and leveled—to avoid cascading inference failures and preserve model performance.
- Delivered with thorough documentation, unit testing, continuous integration, and permissively licensed on PyPI and GitHub.
Jhive
Production (TRL-7)Public • AIforGood
- Django-based web application leveraging the OpenAI API to detect antisemitic content through real-time natural language processing and classification.
- Provides a detailed reporting dashboard for analytical review alongside secure RESTful API endpoints with request rate limiting.
- Implements rigorous input validation and sanitisation measures to prevent injection attacks
- ensuring safe and reliable deployment within sensitive content moderation environments.