Projects
MHIL
Concept & Early Prototyping (TRL-5)Semi-Stealth • Non‑Profit
- MHIL is a non-profit research institute advancing benchmarks and evaluation methods for genuine machine intelligence beyond the limitations of narrow LLM tasks.
- It engineers human-centric test batteries assessing agent capabilities such as memory retention, strategic planning, robustness, and autonomous decision-making under uncertainty.
- The lab develops open evaluation matrices for early 'pre-superintelligent' systems, enabling precise cross-comparison across diverse models, vendors, and architectures.
- Its focus is on setting future-ready standards with reproducible scorecards, tiered capability classifications, and safety-critical thresholds tailored for real-world AI deployment.
MazeByte
SOTA Proof of Concept (PoC)Semi-Stealth
- MazeByte is a pioneering project demonstrating fully autonomous ETL operations through self-maintaining AI agents.
- It incorporates infant-inspired feedback mechanisms involving reward and pain signals alongside resource-seeking behaviours for continual adaptive learning.
- The system autonomously generates pseudo-code modules handling memory, search, optimisation, and orchestrates comprehensive end-to-end ETL automation.
- Leveraging its proprietary MBI model, MazeByte aims to eradicate manual data engineering, transforming traditional data pipelines into self-governing, agentic data systems.
SightBit
SaaS (TRL-9)Public
- SightBit is a production-grade computer vision platform delivering real-time drowning prediction, human and vessel detection, and flood-risk alerting.
- It operates on standard CCTV video streams without reliance on specialised sensors, edge computing devices, or bespoke site tuning.
- Its core uses a panoptic segmentation model enhanced by a proprietary data augmentation pipeline, significantly boosting small-object detection recall by 18 percentage points.
AdaptiveBridge
Production-ready (TRL-8)Public • Open-Source
- AdaptiveBridge is an open-source Python library engineered to maintain machine learning model robustness in production by predicting and imputing missing features at inference time.
- It offers configurable parameters for feature importance, correlations, and supports custom accuracy evaluation alongside automated feature distribution handling.
- The project includes comprehensive documentation, unit tests, continuous integration workflows, and is distributed under a permissive open-source license.
- Integrating seamlessly into existing ML pipelines via a drop-in estimator interface
- it efficiently manages complex feature dependencies including mandatory
Jhive
Production (TRL-7)Public • AIforGood
- Jhive is a Django-based production web application utilising the OpenAI API to detect antisemitic content in real-time.
- It features a detailed content classification and reporting interface designed for operational monitoring and review.
- The system exposes RESTful API endpoints with configurable rate limiting for easy integration into existing workflows.
- Security is paramount, with rigorous input validation and sanitisation processes implemented to prevent injection vulnerabilities and ensure system integrity.